Customer intimacy analytics : leveraging operational data to assess customer knowledge and relationships and to measure their business impact

Habryn, Francois

2012

KIT Scientific Publishing

The ability to capture customer needs and to tailor the provided solutions accordingly, also defined as customer intimacy, has become a significant success factor in the B2B space - in particular for increasingly ""servitizing"" businesses. This book elaborates on the solution CI Analytics to assess and monitor the impact of customer intimacy strategies by leveraging business analytics and social network analysis technology. This solution thereby effectively complements existing CRM solutions.

Business

Management

English

978-3-86644-848-3 (print)

François Habryn
Customer Intimacy Analytics
Leveraging Operational Data to Assess Customer Knowledge
and Relationships and to Measure their Business Impact

Customer Intimacy Analytics
Leveraging Operational Data to Assess Customer
Knowledge and Relationships and to Measure
their Business Impact

by
François Habryn

Dissertation, Karlsruher Institut für Technologie
Fakultät für Wirtschaftswissenschaften,
Tag der mündlichen Prüfung: 16. Februar 2012
Referenten: Prof. Dr. Gerhard Satzger, Prof. Dr. Rudi Studer

Impressum
Karlsruher Institut für Technologie (KIT)
KIT Scientific Publishing
Straße am Forum 2
D-76131 Karlsruhe
www.ksp.kit.edu
KIT – Universität des Landes Baden-Württemberg und
nationales Forschungszentrum in der Helmholtz-Gemeinschaft

Diese Veröffentlichung ist im Internet unter folgender Creative Commons-Lizenz
publiziert: http://creativecommons.org/licenses/by-nc-nd/3.0/de/

KIT Scientific Publishing 2012
Print on Demand
ISBN 978-3-86644-848-3

Customer Intimacy Analytics
Leveraging Operational Data to Assess Customer
Knowledge and Relationships and to Measure
their Business Impact
Zur Erlangung des akademischen Grades
eines Doktors der Wirtschaftswissenschaften
(Dr. rer. pol.)
von der Fakultät für Wirtschaftswissenschaften
des Karlsruher Institut für Technologie (KIT)
genehmigte
Dissertation
von
François Habryn
Tag der mündlichen Prüfung: 16. Februar 2012
Referent:
Prof. Dr. Gerhard Satzger
Korreferent:
Prof. Dr. Rudi Studer
Karlsruhe

Preface
In today’s economies, firms are characterized by an increasing degree of service orientation. Long-term customer relationships and
individualized solutions are emphasized– up to the point where the
willingness and ability to focus on “customer intimacy” turns into a
unique type of strategy, as an alternative to product leadership and
operational excellence (Treacy & Wiersema, 1993). Unfortunately,
currently available CRM systems hardly support customer intimacy
based strategies as they mostly focus on discrete sales transactions.
However, the emerging area of business analytics offers IT-based concepts, methods, and tools that may open up a huge potential for
firms to exploit existing customer interaction data as well as to augment and improve their CRM approach. They actually can provide
tremendous analytical support both for designing and implementing
customer intimacy strategies and for monitoring their effectiveness.
The work of François Habryn takes on this opportunity and makes
significant and innovative contributions along three dimensions. Firstly, it decomposes the notion of “customer intimacy” into operationally
meaningful and measurable components - as a prerequisite for an
analytical evaluation of the quality of customer relationships. Secondly, it develops metrics based on existing interaction data to be
applied to these components. And thirdly it provides a methodology and even a fully-fledged tool to test the ability of these metrics

ii

Preface

to actually reflect relevant customer intimacy in practice. The results
François Habryn obtained in a real case scenario are convincing as
is the positive feedback that he has received at academic conferences
as well as from various industry partners.
The work is a truly remarkable example for the capabilities of interdisciplinary approaches to create innovative solutions to problems:
François Habryn addresses the challenge of assessing customer intimacy from both the managerial and IT perspectives by integrating
concepts grounded in relationship marketing, strategic management,
business analytics, social network analysis, and software engineering. The insights gained should be highly relevant for leaders and
managers in strategy, marketing, and sales in service-oriented companies as well as for consultants and IT providers in the CRM space.
I wish the audience an inspiring, enjoyable, and fruitful reading
of this book and hope that this work will see the distribution in
academia and industry that it deserves.
Prof. Dr. Gerhard Satzger
Director IBM Business Performance Services Europe

Acknowledgements
I would like to express my sincere gratitude to all those who helped
me during the course of this thesis with their advice and support.
First and foremost, I would like to thank Prof. Dr. Gerhard Satzger
for consistently supporting me throughout all the phases of this
project, for always being available to provide me with sound advice,
as well as for leading my research to a high quality. I also wish to
acknowledge the opportunity given to me by IBM in November 2007
to participate in the creation of the Karlsruhe Service Research Institute (KSRI) and to complete this doctoral thesis. This was a fantastic
experience and I am grateful to Gerhard Satzger, Martin Jetter, and
IBM for this opportunity.
I also wish to express my gratitude to Prof. Dr. Rudi Studer for
being the second reviewer of this thesis. His valuable advice, support, and friendly encouragement assisted me greatly in completing
this work. I am also very grateful to Prof. Dr. Hagen Lindstädt and
Prof. Dr. Thomas Lützkendorf for accepting to be part of the examination board as Examiner and Chairman, respectively, and for their
constructive advice and comments.
This work would not have been possible without the support of all
my colleagues at KSRI. I would like to show gratitude in particular
to my colleagues in the Service Innovation and Management team
with whom I spent four excellent years: Prof. Dr. Hansjörg Fromm,

iv

Acknowledgements

Andreas Neus, Robert Kern, Axel Kieninger, Peter Hottum, Marc
Kohler, and Johannes Kunze von Bischhoffshausen. In addition, I
wish to thank Dr. Benjamin Blau, Dr. Arun Anandasivam, Dr. Jeroen
Schepers, and Gielis von der Heijden who advised me in the initial
and final phases of my thesis. I also wish to acknowledge the KIT
students who completed their diploma, bachelor, and master thesis
under my supervision: Thomas Herzig, Lukas Lampe, and Hakan
Bilgic, whose skills and dedication to the project were invaluable.
I would like to show appreciation to CAS Software AG (CAS) with
the help of whom I was able to implement the prototype CI Analytics
and to perform a survey which allowed the overall validation of this
thesis. I would like to express my gratitude to Dr. Bernhard Kölmel
for actively supporting this work within CAS, to Martin Hubschneider for allowing me to perform this project in his company, as well
as to all the CAS employees who participated in the survey.
Finally, I would like to express thanks to those who provided me the
most precious assistance. I owe my deepest gratitude to my parents
for the environment in which I grew up, for their constant encouragement, and for always being there when I needed them. I also wish
to give a very special thank you to Anna for her care and patience.
Anna gave me confidence when I was in doubt and encouraged me
when I was in low spirits.
François Habryn

Abstract
The ability to capture customer needs and to tailor provided solutions accordingly, also defined as customer intimacy, has become a
significant success factor in the Business to Business (B2B) space –
in particular for increasingly “servitizing” businesses. This growing
importance of customer intimacy is driven by a fast development of
the service industry, higher expectations on the demand side, and a
shift in the role of the customer from passive value receiver to active
value co-creator. However, the measurement and management of
customer intimacy lacks analytical support. Even though customer
relationship management (CRM) systems are well established today,
they do not yet provide the appropriate means for supporting the
implementation of a customer intimacy strategy. So far, customer intimacy was not given the adequate focus from the IT perspective and,
thus, many organizations still struggle with measuring and proactively managing the degree of customer intimacy established with
their customers.
In the scope of this thesis, the solution CI Analytics has been conceived, implemented, and validated in order to remedy this issue. CI
Analytics complements existing CRM systems with the capability to
assess and monitor the degree of customer intimacy established by
a provider with its customers in a B2B context. It applies business
analytics and social network analysis technology in order to provide

vi

Abstract

an accurate, real-time, and easily implementable assessment of customer intimacy with two levels of analysis: the individual level and
the organizational level. CI Analytics leverages customer related data
which is available in the information system of the provider (such as
interactions, projects, and sales records) to derive customer intimacy
metrics. These metrics are subsequently used to infer the established
customer intimacy as well as its impact on business results.
Multiple benefits can be derived from the solution proposed by this
thesis. First, CI Analytics allows an organization to benchmark the effectiveness of its customer intimacy strategy with different customers
and, thus, supports this organization with regard to its customer investments. Second, this solution provides a systematic graph-based
overview of the interactions among provider and customer employees, as well as a visualization of their evolution over time, thereby
enabling the provider to proactively act upon any changes in the
activity and interaction patterns with the customer. Finally, CI Analytics fosters the exchange of customer knowledge among the provider employees by facilitating the identification of employees inside
the organization who acquired some specific customer knowledge
and established relationships with customer employees.
The solution CI Analytics has been prototypically implemented in
order to validate the feasibility of the proposed customer intimacy
assessment and monitoring. This software allows different users in
the provider organization to visualize in real time the investments
performed by the provider employees in terms of interaction time
in order to acquire customer knowledge and to establish relationships with customer employees. In addition, this software graphically represents the business impact of the customer intimacy strategy for specific customers and for specific time frames by means
of dedicated customer intimacy performance indicators. CI Analytics
has been evaluated in an enterprise setting with real data from the
IT software and service provider CAS Software AG. This evaluation
confirms the relevance of the proposed solution as well as allows the
organization to gain insights on the patterns of interactions leading
to a successful acquisition of customer knowledge and to an effective
establishment of high-quality customer relationships.

Contents
Preface

i

Acknowledgements

iii

Abstract

v

I.

1

Foundations and Preliminaries

1. Introduction
1.1. Research Problem . . .
1.2. Research Objective . .
1.3. Research Approach . .
1.4. Research Questions . .
1.5. Structure of the Thesis

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

2. Towards Customer Intimacy
2.1. Three Value Disciplines to Achieve Market Leadership
2.1.1. Operational Excellence and Product Leadership
as Alternatives to Customer Intimacy . . . . . .
2.1.2. The Value Discipline Customer Intimacy . . . .
2.2. Customer Intimacy: Grounded in Relationships and
Services . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1. Two Divergent Perspectives on Marketing . . .
2.2.2. The Service Dimension of Relationship Marketing . . . . . . . . . . . . . . . . . . . . . . . . . .

3
6
8
11
14
16
19
20
23
26
31
32
35

viii

Contents

2.2.3. The Service-Dominant Logic as an Evolution of
Relationship Marketing . . . . . . . . . . . . . .
2.2.4. Customer Intimacy: A Relationship and Service Based Value Discipline . . . . . . . . . . . .
2.3. Three Approaches Related to Customer Intimacy . . .
2.3.1. Key Account Management . . . . . . . . . . . .
2.3.2. Market Orientation . . . . . . . . . . . . . . . . .
2.3.3. Customer Relationship Management . . . . . .
2.3.4. Customer Intimacy: A Specific Adoption of the
Marketing Concept . . . . . . . . . . . . . . . . .

38
42
45
45
48
51
55

3. Methods and Techniques to Assess Customer Intimacy
3.1. Network Analysis . . . . . . . . . . . . . . . . . . . . . .
3.1.1. Graph Theory for the Representation of Social
Networks . . . . . . . . . . . . . . . . . . . . . .
3.1.2. Centrality Metrics for the Analysis of Social Networks . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.3. Using Social Network Analysis for Assessing
Customer Intimacy . . . . . . . . . . . . . . . . .
3.2. Data Mining . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.1. The Process of Knowledge Discovery in Databases
3.2.2. Selection of the Machine Learning Algorithms .
3.2.3. Evaluation of the Machine Learning Models . .

59
60

II. Conceptual Model

89

4. Customer Intimacy Breakdown Analysis
4.1. Existing Approaches for Assessing Customer Intimacy
4.2. Overview of the Customer Intimacy Breakdown Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3. Acquired Customer Intimacy Components . . . . . . .
4.3.1. Acquired Customer Knowledge . . . . . . . . .
4.3.2. Established Customer Relationships . . . . . . .
4.4. Leveraged Customer Intimacy Components . . . . . .
4.4.1. Customization . . . . . . . . . . . . . . . . . . .

91
92

61
64
67
69
70
73
82

96
100
101
103
106
107

Contents

ix

4.4.2.
4.4.3.
4.4.4.
4.4.5.
4.4.6.

Loyalty . . . . . . . . . . . . .
Proactiveness . . . . . . . . .
Cross-selling . . . . . . . . . .
Customer Participation . . .
Transaction Costs Reduction

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

109
111
112
114
116

5. CI Analytics Model and Methodology
119
5.1. CI Analytics Overview . . . . . . . . . . . . . . . . . . . 120
5.1.1. CI Analytics Methodology . . . . . . . . . . . . . 120
5.1.2. CI Analytics Model . . . . . . . . . . . . . . . . . 126
5.2. Assessment of the Acquired Customer Intimacy . . . . 129
5.2.1. Using Interactions and Networks to Assess Acquired Customer Intimacy . . . . . . . . . . . . 130
5.2.2. Customer Intimacy Metrics at the Individual
Level . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.2.3. Customer Intimacy Metrics at the Organizational
Level . . . . . . . . . . . . . . . . . . . . . . . . . 147
5.2.4. Empirical Assessment of the Acquired Customer
Intimacy . . . . . . . . . . . . . . . . . . . . . . . 153
5.3. Assessment of the Leveraged Customer Intimacy . . . 156
5.3.1. Customization . . . . . . . . . . . . . . . . . . . 157
5.3.2. Customer Loyalty . . . . . . . . . . . . . . . . . . 158
5.3.3. Proactiveness . . . . . . . . . . . . . . . . . . . . 159
5.3.4. Cross-selling . . . . . . . . . . . . . . . . . . . . . 160
5.3.5. Customer Participation . . . . . . . . . . . . . . 162
5.3.6. Transaction Costs Reduction . . . . . . . . . . . 163

III. Evaluation
6. CI Analytics Software
6.1. CI Analytics Business Analysis .
6.1.1. Requirements Analysis .
6.1.2. Business Objects Analysis
6.2. CI Analytics Architecture . . . . .
6.2.1. Architecture Overview . .

165
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

167
168
168
176
179
179

x

Contents

6.2.2. CI Data Warehouse . . . . . .
6.2.3. CI ETL . . . . . . . . . . . . .
6.2.4. CI Services . . . . . . . . . . .
6.2.5. CI Dashboard . . . . . . . . .
6.3. CI Analytics Evaluation . . . . . . . .
6.3.1. Requirements Assessment . .
6.3.2. Business Benefits Evaluation

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

181
185
187
191
196
197
202

7. CI Analytics Validation
211
7.1. Acquired Customer Intimacy at the Individual Level . 213
7.1.1. Data Collection . . . . . . . . . . . . . . . . . . . 213
7.1.2. Calibration: Acquired Knowledge . . . . . . . . . 221
7.1.3. Calibration: Established Relationships . . . . . . . 234
7.2. Acquired Customer Intimacy at the Organizational Level242
7.2.1. Data Collection . . . . . . . . . . . . . . . . . . . 243
7.2.2. Calibration: Acquired Knowledge . . . . . . . . . 247
7.2.3. Calibration: Established Relationships . . . . . . . 255
7.3. Summary and Interpretation of the Calibration Results 262
7.3.1. Results Summary . . . . . . . . . . . . . . . . . . 262
7.3.2. Results Interpretation . . . . . . . . . . . . . . . 266
8. Conclusion
269
8.1. Contribution . . . . . . . . . . . . . . . . . . . . . . . . . 270
8.2. Managerial Implications . . . . . . . . . . . . . . . . . . 275
8.3. Outlook on Future Research . . . . . . . . . . . . . . . . 278
Bibliography

285

Appendix
307
A. Questionnaire Customer Intimacy . . . . . . . . . . . . 307
A.1. English Version . . . . . . . . . . . . . . . . . . . 307
A.2. German Version . . . . . . . . . . . . . . . . . . . 312
B. Machine Learning Algorithms Settings . . . . . . . . . 317
C. Acquired Customer Intimacy at the Individual Level . 322
D. Acquired Customer Intimacy at the Organizational Level333

Contents

E.

xi

CI Analytics Implementation . . . . . . . . . . . . . . .
E.1.
CI Services for Calculating the Acquired Customer Intimacy Metrics . . . . . . . . . . . . . .
E.2.
CI Services for Calculating the Leveraged Customer Intimacy Metrics . . . . . . . . . . . . . .
E.3.
CI Graph: A First Prototype of CI Analytics . . .
E.4.
Business Benefits Analysis . . . . . . . . . . . .

342
342
345
348
353

List of Figures
1.1. Different Degrees of Customer Intimacy Between Provider and Customer Entities . . . . . . . . . . . . . . . .
1.2. Structure of the Thesis . . . . . . . . . . . . . . . . . . .

10
17

2.1. Three Value Disciplines to Achieve Market Leadership
2.2. Customer Intimacy Operating Model . . . . . . . . . .
2.3. Exchange and Relationship Perspectives . . . . . . . . .

21
29
34

3.1. A Weighted Bipartite Graph Representation of the ProviderCustomer Relationship . . . . . . . . . . . . . . . . . . . 64
3.2. The Knowledge Discovery Process . . . . . . . . . . . . 71
3.3. Illustrative Multilayer Perceptron . . . . . . . . . . . . . 81
3.4. Confusion Matrix . . . . . . . . . . . . . . . . . . . . . . 85
3.5. ROC Curve . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.1. The Two Dimensions of Customer Intimacy . . . . . . . 98
4.2. Breakdown Analysis of the Acquired and Leveraged
Customer Intimacy . . . . . . . . . . . . . . . . . . . . . 100
CI Analytics Methodology . . . . . . . . . . . . . . . . . 122
CI Analytics Model . . . . . . . . . . . . . . . . . . . . . 127
Interaction Levels in a Relationship . . . . . . . . . . . 133
Customer Interaction Time: A Means To Aggregate
Customer Interaction Across Multiple Channels . . . . 136
5.5. Segmentation of the Relationship to Identify Episodes
Across Multiple Channels . . . . . . . . . . . . . . . . . 140
5.6. Two Different Graph Representations of the Social Network Formed by the Provider and Customer Employees 146
5.1.
5.2.
5.3.
5.4.

xiv

List of Figures

CI Analytics Architecture . . . . . . . . . . . . . . . . . .
Customer Interaction Time Star Schema . . . . . . . . .
Overview of the CI ETL Process . . . . . . . . . . . . .
Main Interface of the CI Dashboard . . . . . . . . . . .
CI Dashboard: Acquired Customer Intimacy . . . . . .
CI Dashboard: Leveraged Customer Intimacy . . . . .
CI Analytics: Business Benefit 1 – Question 5 . . . . . .
CI Analytics: Business Benefit 1 – Question 6 . . . . . .
CI Analytics: Business Benefit 2 – Question 7 . . . . . .
CI Analytics: Business Benefit 2 – Question 8 . . . . . .
CI Analytics: Business Benefit 3 – Question 9 . . . . . .
CI Analytics: Business Benefit 3 – Question 10 . . . . .
CI Analytics: Overall Appreciation and Data Privacy
Concerns . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.14. CI Analytics: Overall Appreciation and Data Privacy
Concerns . . . . . . . . . . . . . . . . . . . . . . . . . . .

181
183
187
192
195
196
204
204
205
206
207
207

7.1. Creation of the Calibration Data Set . . . . . . . . . . .
7.2. Knowledge High: Decision Tree Model and ROC Curve
7.3. Knowledge Very High: Decision Tree Model and k-nearest
Neighbor ROC Curve . . . . . . . . . . . . . . . . . . .
7.4. Relationship High: Decision Tree Model and k-nearest
Neighbor ROC Curve . . . . . . . . . . . . . . . . . . .
7.5. Relationship Very High: Decision Tree Model and ROC
Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.6. Knowledge High: Decision Tree Model and ROC Curve
7.7. Knowledge Very High: Decision Tree Model and ROC
Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.8. Relationship High: Decision Tree Model and Multilayer
Perceptron ROC Curve . . . . . . . . . . . . . . . . . . .
7.9. Relationship Very High: Decision Tree Model and ROC
Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

218
231

6.1.
6.2.
6.3.
6.4.
6.5.
6.6.
6.7.
6.8.
6.9.
6.10.
6.11.
6.12.
6.13.

208
208

234
239
241
253
254
259
261

A.1. Customer Intimacy Questionnaire: Introduction . . . . 308
A.2. Customer Intimacy Questionnaire: Acquired Customer
Intimacy at the Organizational Level . . . . . . . . . . . 309

List of Figures

A.3. Customer Intimacy Questionnaire: Acquired Customer
Intimacy at the Individual Level . . . . . . . . . . . . .
A.4. Customer Intimacy Questionnaire: Work Environment
A.5. Customer Intimacy Questionnaire: Introduction (German) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.6. Customer Intimacy Questionnaire: Acquired Customer
Intimacy at the Organizational Level (German) . . . . .
A.7. Customer Intimacy Questionnaire: Acquired Customer
Intimacy at the Individual Level (German) . . . . . . .
A.8. Customer Intimacy Questionnaire: Work Environment
(German) . . . . . . . . . . . . . . . . . . . . . . . . . . .
C.1. Crombach’s Alpha of the Scales Knowledge and Relationship at the Individual Level . . . . . . . . . . . . . .
D.1. Crombach’s Alpha of the Scales Knowledge and Relationship at the Organizational Level . . . . . . . . . . . .
E.2. CI Graph: Architecture Overview . . . . . . . . . . . . .
E.3. CI Graph: Calibration Panel . . . . . . . . . . . . . . . .
E.4. CI Graph: Visualization Panel . . . . . . . . . . . . . . .
E.5. CI Analytics: Business Benefits Questionnaire (1/3) . .
E.6. CI Analytics: Business Benefits Questionnaire (2/3) . .
E.7. CI Analytics: Business Benefits Questionnaire (3/3) . .
E.8. Business Benefits Survey: Participants Profiles . . . . .

xv

310
311
313
314
315
316
322
333
349
351
352
354
355
356
356

List of Tables
2.1. Comparison of Customer Intimacy With Other Marketing Programs . . . . . . . . . . . . . . . . . . . . . . . .

56

4.1. Overview of Existing Approaches Towards the Assessment of Customer Intimacy . . . . . . . . . . . . . . . .

94

5.1. Customer Intimacy Metrics at the Individual and Organizational Levels . . . . . . . . . . . . . . . . . . . . . 154
5.2. Customer Intimacy Metrics for the Leveraged Customer
Intimacy . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
6.1. Functional and Non-Functional Requirements on CI
Analytics . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.2. CI Analytics Business Objects . . . . . . . . . . . . . . .
6.3. CI Services Overview . . . . . . . . . . . . . . . . . . . .
6.3. CI Services Overview (Continued) . . . . . . . . . . . . .
6.4. Fulfillment of the Functional and Non-Functional Requirements . . . . . . . . . . . . . . . . . . . . . . . . . .
7.1. Model Configurations and Metrics to Assess Acquired
Customer Intimacy at the Individual Level . . . . . . .
7.2. Creation of the Calibration Data Set . . . . . . . . . . .
7.3. Proportions of Knowledge High and Knowledge Very High
Records . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.4. Proposed Interpretation of the Performance Indicators
7.5. Knowledge High: Performance Indicator Results . . . . .
7.6. Knowledge Very High: Performance Indicator Results . .
7.7. Proportions of Records of Class Relationship High and
Relationship Very High . . . . . . . . . . . . . . . . . . . .

169
177
190
191
197
214
220
226
228
230
232
237

xviii

List of Tables

7.8. Relationship High: Performance Indicator Results . . . .
7.9. Relationship Very High: Performance Indicator Results .
7.10. Model Configurations and Metrics to Assess Acquired
Customer Intimacy at the Organizational Level . . . .
7.11. Proportions of Records of Class Knowledge High and
Knowledge Very High . . . . . . . . . . . . . . . . . . . .
7.12. Proposed Interpretation of the Performance Indicators
7.13. Knowledge High: Performance Indicator Results . . . . .
7.14. Knowledge Very High: Performance Indicator Results . .
7.15. Proportions of Relationship High and Relationship Very
High Records . . . . . . . . . . . . . . . . . . . . . . . . .
7.16. Relationship High: Performance Indicator Results . . . .
7.17. Relationship Very High: Performance Indicator Results .
7.18. Summary of the Calibration Results . . . . . . . . . . .
7.19. Number of Occurrences of the Metrics in the Decision
Tree Models . . . . . . . . . . . . . . . . . . . . . . . . .

238
240
244
250
251
252
254
258
259
261
263
265

B.1. Configuration Settings of the Decision Tree C4.5 . . . . 318
B.2. Configuration Settings of the k-nearest Neighbor Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
B.3. Configuration Settings of the Support Vector Machine
Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . 319
B.4. Configuration Settings of the Multilayer Perceptron with
Backpropagration . . . . . . . . . . . . . . . . . . . . . . 320
B.5. Number of Tested Configurations of the Machine Learning Algorithms to Predict the Customer Intimacy Values321
C.1. Prediction of the Variable Knowledge High: Detailed
Performance Results of the Decision Tree C4.5 . . . . . 323
C.2. Prediction of the Variable Knowledge High at the Individual Level: Best Configurations and Results . . . . . 325
C.3. Prediction of the Variable Knowledge Very High at the
Individual Level: Best Configurations and Results . . . 327
C.4. Prediction of the Variable Relationship High at the Individual Level: Best Configurations and Results . . . . . 329
C.5. Prediction of the Variable Relationship Very High at the
Individual Level: Best Configurations and Results . . . 331

List of Tables

xix

D.1. Prediction of the Variable Knowledge High at the Organizational Level: Best Configurations and Results . . . 334
D.2. Prediction of the Variable Knowledge Very High at the
Organizational Level: Best Configurations and Results 336
D.3. Prediction of the Variable Relationship High at the Organizational Level: Best Configurations and Results . . 338
D.4. Prediction of the Variable Relationship Very High at the
Organizational Level: Best Configurations and Results 340
E.1. CI Services For Calculating the Acquired Customer Intimacy Metrics: Technical Details . . . . . . . . . . . . . 344
E.2. CI Services for the Leveraged Customer Intimacy Metrics345

Part I.

Foundations and
Preliminaries

1. Introduction
A recent survey conducted in 2010 with 1500 chief executive officers
worldwide established that, today, successful organizations “make
customer intimacy their number-one priority” (IBM Institute for Business Value, 2010, p.9). Customer intimacy has gained momentum
over the last years as it is perceived as a means to develop a sustainable business strategy in mature markets such as Europe and
the United States, which are characterized by limited growth, fierce
competition, and demanding customers.
Customer intimacy was first introduced by Treacy & Wiersema (1993)
as one of three value disciplines, along with operational excellence
and product leadership that, if executed well, leads to market leadership. Several firms in various industries, including IBM, Panalpina,
Unilever, and General Electric Healthcare, were influenced by the
concept of customer intimacy. Defined as the ability to “continuously tailor and shape products and services to fit an increasingly
fine definition of the customer” (Treacy & Wiersema, 1993, p.87), customer intimacy determines an organization’s business strategy and,
as such, critically impacts its operations and performance (Hambrick,
1980).
In the contemporary challenging business environment, multiple companies strive to find new sources of growth and, thus, strengthen
their relationships with customers in order to achieve new forms

4

1. Introduction

of competitive advantages (Tuominen et al., 2004; Day, 2003). In
that regard, a firm which successfully pursues customer intimacy
derives strategic benefits from its knowledge of, and relationship
with, customers. For instance, the customer intimate firm proactively improves its value proposition and becomes its customers’ preferred partner by customizing its offering to their specific requirements (Wallenburg, 2009). This firm embeds its customers in the
value creation process and leverages their ideas in order to conceive
innovative solutions. The firm adhering to customer intimacy also
increases the loyalty of its customers, thereby protecting its investments by establishing long-term relationships.
Customer intimacy is particularly relevant in a service context as the
development of a customer intimacy strategy covers two essential
characteristics of services, namely the individualization of the offering to customer needs, and the intensification of the customer interactions in order to co-create value with customers (Bruhn & Georgi,
2006). Numerous companies that were known for their product centered portfolios have developed services-focused business models.
For instance, Rolls-Royce and IBM, which used to generate over 60%
of their revenues in 1995 with products, redesigned their offerings
and realized in the past three years over 55% of their revenues with
services. This transformation of the firm’s business model from selling goods to selling solutions including goods and services is called
servitization (Vandermerwe, 1988; Neely, 2009). Customer intimacy
is potentially an adequate type of business strategy for companies
undergoing a servitization endeavor and which try to strengthen
their relationships with customers and to individualize their offerings.
Pursuing a customer intimacy strategy poses some specific challenges
to the organization. In order to implement a customer intimacy strategy, the provider needs to manage the relationships established with
customers as well as the acquired knowledge related to customers. In
that regard, this thesis focuses on the specific challenges of business
to business (B2B) markets. In a B2B context, both the provider and
the customer consist of multiple teams and individuals. On the customer side, this means that, in most cases, users and purchasers are

5

different individuals inside the customer organization. While the decision to select one or the other B2B provider is made by purchasers,
the users actually get in contact with the provided solution, and as
such assess its quality and performance. Thus, the B2B provider
must consider the needs and requirements of the different stakeholders inside the customer organization in order to successfully manage
the relationship with the customer organization and successfully implement its customer intimacy strategy (Homburg & Jensen, 2004).
On the provider side, the ongoing servitization has a substantial
impact on the organization, blurring the boundaries between sales,
services, marketing, and even manufacturing departments (Oliva &
Kallenberg, 2003). Sales employees, who were spokespersons for the
firm’s products have become sales consultants who understand and
solve customer problems, leveraging knowledge and expertise across
the entire provider organization (Sheth & Sharma, 2008). Reciprocally, service employees become increasingly involved in the selling
process as they develop unique means to gather customer knowledge and understand the customer’s mindset. Thus, managing customer relationships and pursuing customer intimacy in a B2B context requires the provider to thoroughly manage the complex and
dynamic social network resulting from the interactions of his employees with customer ones. This development drives the need to redesign the interfaces among the internal departments of the B2B
provider, “in terms of structure, communication patterns, information sharing, collaboration, and strategic outcome” (Biemans et al.,
2010, p.183). Zack et al. (2009, p.402) confirm that “firms achieving
high customer intimacy engaged in the widest range of knowledge
management practices.”
From an academic perspective, customer intimacy overlays, in part,
with prominent marketing concepts such as relationship marketing
(Berry, 1983) and the modern perspective on services, namely the
service-dominant logic (Vargo & Lusch, 2004a). Relationship marketing and the service-dominant logic take their root in a paradigm
shift that positions the relationship with the customer as a central
determinant of the marketing strategy, rather than the delivery of
the product or service itself. Grönroos (1994), for instance, contrasts

6

1. Introduction

the “4Ps marketing mix” of the transactional marketing which is
dominated by the quality of the output and measured by market
share, with relationship marketing, which is driven by the quality
of the customer interactions and individually measured with each
customer. Vargo & Lusch (2008b) qualify the service-dominant logic
as focused on the exchange of knowledge and skills among partners
rather than on the exchange of tangible goods, thereby contrasting
the service-dominant logic with the goods-dominant logic. As it will
be explained in chapter 2, customer intimacy is rooted in the concept
of relationship marketing and shares several commonalities with the
service-dominant logic.

1.1. Research Problem
From the IT perspective, the choice to pursue the value discipline
customer intimacy directly impacts the IT governance of the organization and its infrastructure design. Weill & Ross (2004) investigated the influence of customer intimacy on IT governance by means
of a survey with 250 enterprises worldwide. They concluded that
customer intimacy driven organizations “strive for a single view of
the customer”, require analytical tools “to expose customers with
the greatest lifetime value”, and “implement customer relationship
management (CRM) systems to support data standardization” (Weill
& Ross, 2004, p.164). CRM systems aim at enabling to collect vast
amounts of customer data and to constructively analyze, interpret,
and utilize it (Payne, 2005). Such systems, therefore, support the development of a customer intimacy strategy. Several sources confirm
that CRM systems have been widely adopted in order to achieve this
objective. A recent Gartner report estimates the size of the CRM application market over $10 billion (Maoz et al., 2010). Sackmann et al.
(2008) found that, in 2008, 68% of the 292 German enterprises they
surveyed had already implemented a CRM solution, and another
20% were planning to do so in 2009.
However, some evidence leads to question the actual benefits of CRM
systems and in particular their positive association with business
performance (Reinartz et al., 2004). While Kale (2004) estimated the

1.1. Research Problem

7

CRM project failure rate between 60% and 80%, Dickie (2007) evaluated that only 20% of the organizations generated additional revenues from their CRM investments. Even though the customization of
products and services is established as a value driver for the adoption of CRM, several CRM projects solely lead to an improvement of
sales force efficiency and effectiveness (Richards & Jones, 2008). Blois
(2008, p.1) states that “(CRM) software on the market today helps automate processes, but does not necessarily provide incremental value
back to the user.” He also considers that CRM systems are only used
to track the progression of the sales opportunities from initial leads
towards contracts (Blois, 2008).
In order to explain this phenomenon, Liang (2009) considers that
IT systems have been so far adopted with transactional focus and
operational excellence in mind. He argues that “the role of customer intimacy has been under-investigated” from the IT perspective (Liang, 2009, p.1). Even though CRM systems aim at managing customer related data, the customer knowledge which is derived
from this data is mostly limited to the transactional perspective. The
CRM system helps answering questions such as which products have
been sold, in which quantities, when and by whom. However, more
complex questions related to the needs of the customer, his future
plans, or his purchasing behavior hardly find an answer in such systems. A survey performed with 122 senior executives in Western Europe acknowledges that firms’ knowledge management capabilities
are the weakest when knowledge is related to customers: “Despite
the heavy investments firms have made in CRM systems in recent
times, only 23% of the surveyed executives say they are effective in
capturing and exploiting information on customer preferences and
behavior” (Ernerst-Jones, 2005, p.7).
Considering the employees’ perspective, this survey also indicates
that organizations particularly struggle with exploiting knowledge of
their employees (Ernerst-Jones, 2005). It is most likely that some provider employees who have spent time working for, and interacting
with, the customer know the customer processes, how decisions are
influenced and taken, and how budget is made available in the customer organization. These employees know how to effectively bring

8

1. Introduction

new ideas inside the customer organization and, reciprocally, how
to obtain useful feedback from the customer. They are also aware of
the customer employees that favor their own organization and those
who favor the competitors. In short, these provider employees know
how to manage the three types of customer knowledge proposed
by Gibbert et al. (2002): about the customer, from the customer, and
for the customer. Thus, these employees have developed a certain
degree of customer intimacy with the customer and the customer
employees. However, because customer knowledge is often tacit and
quickly outdated, provider employees do not have the means to store
it in an explicit manner in the CRM system, and the provider does
not have the capability to assess the degree of customer intimacy
established by its employees with its different customers.
At the organizational level, the customer contribution margin is the
most basic conception for assessing the profitability of business relationships (Wengler, 2006). However, an empirical analysis performed
in 2006 reveals that only 30% of the surveyed organizations take this
parameter into account, and 80% of them solely use transaction volumes in order to rate their customers (Wengler et al., 2006). Taking
the broader perspective of customer intimacy, a thorough literature
review (Habryn et al., 2010) which is further refined in section 4.1
of this thesis acknowledges that, as of today, there is no operational
means for an organization to assess the degree of customer intimacy
established with customers.
As a result, the central problem which is investigated in the scope of
this thesis is concerned with the lack of easily exploitable solutions
for an organization to assess the degree of customer intimacy that it
has established at both the individual and organizational levels with
its customers.

1.2. Research Objective
In order to address the issue presented in the previous section, the
objective of this thesis is to develop a solution for assessing and monitoring the degree of customer intimacy established by an organization with its customers.

1.2. Research Objective

9

As illustrated in figure 1.1, the various interactions and activities of
the provider employees with customer employees lead to the establishment of different degrees of customer intimacy between entities
of the provider and customer. For instance, it is most likely that provider employees who worked on a customer project at the customer
location developed a higher customer intimacy than other employees
who only had limited interactions with the customer: they gathered
more knowledge about the customers as they spent time with its
employees and used this knowledge to adapt the solution they developed. The different business units, teams, and employees of the provider, thus, established different degrees of customer intimacy with
the business units, teams, and employees of the customer. In order to
analyze the degree of customer intimacy between the provider and
the customer, it is therefore necessary to drill down the analysis to
multiple levels of details.
Consequently, the assessment of the degree of customer intimacy
should be performed in the scope of this thesis at two levels of granularity: the organizational level and the individual level. The organizational level indicates the customer intimacy established with customer organizations and its entities such as teams and business units.
The individual level refers to the degree of customer intimacy established by provider employees with customer employees.
In order for this customer intimacy assessment to be relevant and usable by a provider, it needs to be up-to-date, accurate, and easily implementable. Making up-to-date assessments is a particularly challenging task as the information related to customer intimacy changes
rapidly. For instance, customer needs may quickly evolve after a
strategic reorientation. The customer might change its purchasing
policy or decide to develop a new market for which he has new requirements. In addition, the customer organization and structure are
also modified on a regular basis. If some customer employees with
whom the provider had established qualitative relationships take a
new position, the provider’s ability to access knowledge about the
customer and influence the customer may decrease, thereby impacting the customer intimacy established with the customer. The pro-

10

1. Introduction

Provider

Bus.
Unit
2

Customer

C

Team
2

P

Team
z

Bus.
Unit C

R

B
Team
3

Bus.
Unit
B

Q

A

Bus.
Unit
2

Team
y

Team 1

Team x

Business Unit 1

Business Unit A

Customer
Intimacy

Figure 1.1.: Different Degrees of Customer Intimacy Between Provider and Customer Entities
vider must, therefore, have up-to-date information on such changes
in order to successfully implement its customer intimacy strategy.
In order to obtain accurate information, this degree of customer intimacy should be evaluated across all departments of the provider
organization. Indeed, the lack of information for specific provider
employees or teams might lead to wrong interpretation of the degree
of customer intimacy and restrain the ability to disseminate customer
knowledge inside the provider organization. If an provider employee
who has a very strong insight about the customer is not identified,
his colleagues cannot benefit from his knowledge.
Finally, to achieve an easily implementable solution, the approach
should not impact the provider employees with significant additional
workload and should integrate seamlessly with the existing IT environment. If the provider employees have to spend a lot of time to
enter customer intimacy related data into the system, they will be
reluctant to using this solution. For these reasons, the customer intimacy assessment should be performed as far as possible automati-

1.3. Research Approach

11

cally, in a real-time fashion, and it should leverage readily available
data.

1.3. Research Approach
In order to fulfill the requirements outlined in the previous section,
this thesis is grounded in the areas of business intelligence and business analytics. The notion of business intelligence has been given
multiple meanings in past literature. Turban et al. (2011, p.19) suggest a broad interpretation of business intelligence and define it as
“an umbrella term that combines architectures, tools, databases, analytical tools, applications, and methodologies [...] to enable interactive access to data, to enable manipulation of data, and to give [...] the
ability to conduct appropriate analysis.” According to Turban et al.
(2011), business analytics is a part of business intelligence which is
explicitly concerned with the exploitation of data by business users
by means of either simple reports and queries or sophisticated mathematical and statistical methods such as data-mining. Davenport &
Harris (2007, p.7) acknowledge that analytics is a subset of business
intelligence and define it as “the extensive use of data, statistical and
quantitative analysis, explanatory and predictive models, and factbased management to drive decisions and actions.”
Business intelligence and in particular business analytics have received a growing interest over the past years. They are perceived as
the means to take informed decisions upon the vast amount of gathered data and as a new source of competitive advantages (Davenport,
2006). An extensive study performed with 4500 managers and executives acknowledges that “58% of organizations now apply analytics
to create a competitive advantage within their markets or industries,
up from 37 percent just one year ago” (Kiron et al., 2011). In addition,
this study confirms the relevance of business analytics for supporting the development of a customer intimacy strategy, as 62% of the
organizations having a strong and sophisticated usage of analytics already leverage analytics for creating personalized relationships with
customers. The importance of business analytics is also confirmed
in a global survey conducted with 1700 chief marketing officers in

12

1. Introduction

which 81% of the respondents confirmed their intent to use business
analytics solutions over the next three to five years (IBM Institute for
Business Value, 2011, p.26).
Following a business intelligence approach, the solution CI Analytics
proposed by this thesis builds upon the idea that customer related
data which is stored inside the information system of the provider,
such as interactions, activities, projects, and results data contains evidence of customer intimacy. Thus, this thesis aims at finding a set
of metrics which can be calculated upon customer related data and
which enables the assessment and monitoring of the degree of customer intimacy established by the provider with its customers at both
the individual and organizational levels. As suggested by De Choudhury et al. (2010), two problems have to be considered: the inference
and the relevance of the customer intimacy metrics. The inference issue relates to the fact that the customer intimacy components are not
directly observable and need to be inferred out of existing customer
data. For instance, even though previous research indicates a positive association between interactions, customer knowledge, and customer relationships (Ballantyne, 2004), there is no rule establishing
that a specific frequency or duration of customer interaction leads
to acquiring customer knowledge or establishing customer relationships. The relevance of the customer intimacy metrics is the second
main challenge of the customer intimacy assessment proposed by
this thesis. An infinite number of metrics can potentially be calculated out of the existing customer data. The challenge is, therefore,
to perform the best selection of customer intimacy metrics in order
to accurately assess the customer intimacy components. The customer intimacy metrics have to be sorted and weighted according
to their relevance for performing this assessment. The CI Analytics
methodology which is proposed in chapter 5 aims at solving these
two challenges of inference and relevance of the customer intimacy
metrics.
Focusing on the inference challenge, a central aspect of this thesis
and of the solution CI Analytics relates to the identification of interaction patterns indicating the development of customer relationships
and the acquisition of customer knowledge, and from which some

1.3. Research Approach

13

customer intimacy metrics can be derived. This thesis, thus, intends
to provide an innovative contribution to the business analytics subset
called interaction analytics which has been qualified by Gartner as a
technology trigger in the hype cycle for analytics applications (Gartner, 2010). Another key aspect of the solution CI Analytics is concerned with the determination of results oriented metrics allowing
the assessment of the business impact of the customer intimacy strategy at the organizational level. In that regard, CI Analytics relates
to the discipline of pattern-based strategies which is defined as the
search for “patterns that may have a positive or negative impact on
business strategy and operations” (Burton et al., 2011).
In order to perform the customer intimacy assessment, this thesis
also relies on network analysis methods (Brandes & Erlebach, 2005b).
Such methods have already been successfully applied for assessing
relationships in B2B context (Gummesson, 2008, p.296), and previous
research already proved that the effectiveness of key account management is affected by the properties of the social network formed
by the provider and customer employees, such as the size of the
network and the position of the employees in the network (Hutt
& Walker, 2006). The solution CI Analytics proposed by this thesis
therefore provides a graph-based representation of the customer intimacy information and uses network topology metrics such as the
degree and closeness centralities as input for the customer intimacy
assessment.1
The solution CI Analytics proposed by this thesis yields the following
benefits:
• First, CI Analytics provides a systematic graph-based overview
of the interactions among provider and customer employees, as
well as a visualization of their evolution over time. A change
in the interaction pattern can be identified, thereby allowing
the provider to proactively act upon it. For instance, frequent
interactions with the support team could indicate customer dissatisfaction. A drop in the interaction between two employees
could mean that the customer organization has been modified.
1

Further details are provided in chapter 5.

14

1. Introduction

In both cases, some actions should be taken by the provider on
the basis of this information.
• Second, this approach fosters the exchange of customer knowledge among the provider employees. CI Analytics enables the
identification of the provider employees who own customer related knowledge and who have established relationships with
the customer. By making this information available in the form
of a graph representation, provider employees having and seeking customer knowledge can identify each other and share tacit
customer knowledge. For instance, a service employee who
knows the customer could inform his colleague working in
sales about the best ways to approach the customer and provide meaningful insights on the customer needs.
• Finally, the solution CI Analytics provides the ability to benchmark the effectiveness of the customer intimacy strategy with
different customers. The provider can identify to which extent
the customer intimacy strategy was executed with each customer and identify which customers are responsive to the customer intimacy strategy. For instance, if the provider invests
resources in adapting the solution proposed to the customer,
but the customer disregards this solution and selects a cheaper
one, then the provider should consider changing its strategy
with this customer. CI Analytics allows the identification of such
patterns out of customer related data.

1.4. Research Questions
The three concrete research questions addressed in the scope of this
thesis are derived from the previously outlined research objective.
Research Question 1 – How can the concept of customer intimacy
be broken down into multiple assessable customer intimacy components?
Customer intimacy is a complex type of strategy which includes multiple facets. In order to perform the assessment of the degree of

1.4. Research Questions

15

customer intimacy achieved by a provider with its customers, a thorough understanding of the concept of customer intimacy is required.
The first research question of this thesis is therefore concerned with
the analysis and identification of the key components of the customer
intimacy strategy. With the original definition of customer intimacy
provided by Treacy & Wiersema (1993) as the starting point of the
analysis, this thesis derives multiple measurable and actionable customer intimacy components, thereby creating the foundations of the
customer intimacy model proposed in chapter 5.
Research Question 2 – Which metrics can be created upon customer
related data in order to infer the customer intimacy components?
The second research question of this thesis relates to the conception
of customer intimacy metrics which can be calculated upon customer
data stored in the information system of the provider. These metrics
should provide the means to determine the values of the customer intimacy components. This question, thus, relates to the previously introduced challenge of inferencing the customer intimacy components
upon customer intimacy metrics. Leveraging literature grounded in
the fields of interaction and relationship marketing, some interaction,
activity, and result patterns are identified and used to conceive significant customer intimacy metrics. The CI Analytics model which is
proposed in chapter 5 elaborates on the customer intimacy metrics
proposed by this thesis to infer the values of the customer intimacy
components. For validation purposes, this model has been embedded in the software CI Analytics which was implemented in the scope
of this thesis. This software described in chapter 6.
Research Question 3 – Which combination of metrics provides the
most accurate assessment of the customer intimacy components?
The third research question is concerned with the determination of
the relative importance of the customer intimacy metrics for accurately assessing the values of the customer intimacy components.
This question therefore relates to the previously mentioned challenge
of relevance of the customer intimacy metrics. Since each provider

16

1. Introduction

manages the relationship with its customers and interacts with the
customer employees in a specific way, the relevance of the customer
intimacy metrics is influenced by the specific activity and interaction
patterns of the provider: some metrics may be relevant for a specific provider and irrelevant for another one. In order to answer this
research question, this thesis proposes in chapter 5 the CI Analytics
methodology for determining the relevance of the customer intimacy
metrics and for calibrating them to the activity and interaction patterns of the provider. This methodology is based on data-mining and
on machine learning algorithms which are explained in chapter 3. In
order to validate the results of this thesis, the CI Analytics methodology has been tested in a real-case scenario with the IT software and
service provider CAS Software AG.2 The results of this validation are
proposed in chapter 7.

1.5. Structure of the Thesis
As depicted in figure 1.2, this thesis is structured into three parts and
eight chapters.
Part 1 presents the foundational and preliminary knowledge which
is required in order to understand this thesis.
• Chapter 1 (Introduction) defines the context of this thesis, details the research problem, and sets out the research questions
addressed by this thesis.
• Chapter 2 (Towards Customer Intimacy) elaborates on the concept
of customer intimacy as defined in past literature and analyzes
its distinctive characteristics with regard to other prominent
marketing concepts such as relational marketing, the servicedominant logic, and key account management.
• Chapter 3 (Methods and Techniques to Assess Customer Intimacy)
lays down the methods and techniques leveraged by this thesis
to achieve the objective of assessing customer intimacy upon
2

Further information on CAS are available at www.cas.de (accessed on
10.11.2011).

1.5. Structure of the Thesis

Part 1
Foundations and
Preliminaries

Part 2
Conceptual
Model

Part 3
Evaluation

17

Chapter 1
Introduction
Chapter 2
Towards Customer Intimacy

Chapter 3
Methods and Techniques to
Assess Customer Intimacy

Chapter 4
Customer Intimacy Breakdown Analysis
Chapter 5
CI Analytics Model and Methodology

Chapter 7
CI Analytics Validation

Chapter 6
CI Analytics Software
Chapter 8
Conclusion

Figure 1.2.: Structure of the Thesis
customer related data available in the provider’s information
system. More specifically, this chapter introduces fundamental
knowledge on graph theory and social network analysis as well
as on data mining and machine learning algorithms.
Part 2 elaborates on the conceptual model proposed by this thesis
and establishes the means and methodology allowing the assessment
of customer intimacy in a B2B context.
• Chapter 4 (Customer Intimacy Breakdown Analysis) analyzes the
concept of customer intimacy and establishes how it can be broken down in multiple quantifiable components. This chapter,
thus, sets out the foundation of the overall model to assessing
and monitoring customer intimacy.
• Chapter 5 (CI Analytics Model and Methodology) elaborates on the
CI Analytics model proposed by this thesis to assess customer
intimacy and develops a set of metrics allowing the measurement of the customer intimacy components defined in chap-

18

1. Introduction

ter 4. This chapter also develops the CI Analytics methodology
conceived to use the CI Analytics model as well as to calibrate it
to the specific patterns of the organization adopting the model.
Part 3 provides an evaluation of the proposed CI Analytics model and
methodology elaborated in chapters 4 and 5.
• Chapter 6 (CI Analytics Software) details the software CI Analytics which has been conceived in the scope of this thesis to
implement the CI Analytics model and to calculate the customer
intimacy metrics proposed in chapter 5. This chapter, thus, validates the feasibility of the customer intimacy assessment proposed by this thesis.
• Chapter 7 (CI Analytics Validation) develops a real case scenario
with the IT software and service provider CAS Software AG in
which the CI Analytics methodology proposed in chapter 5 has
been applied. Following this methodology, this chapter depicts
how the customer intimacy metrics have been calculated upon
real data related to 14 different customers and calibrated to fit
the characteristics of this provider. This chapter subsequently
presents the results of the evaluation of this calibration, thereby
validating the overall approach of this thesis to assessing customer intimacy.
• Chapter 8 (Conclusion) analyzes the extent to which the research
questions defined in chapter 1 have been answered by this thesis and summarizes its contribution. This chapter subsequently
outlines the managerial implications which can be derived from
this thesis, addresses the limitations of its findings, and suggests directions for future research.

2. Towards Customer Intimacy
Several aspects are recurrent when reading literature on customer
intimacy such as gaining a competitive advantage, developing a strategy, or managing customer knowledge and relationships. In order
to achieve the goal of this thesis to assess customer intimacy, it is
therefore necessary to fully understand this concept as well as these
different aspects. Moreover, during and after the concept of customer intimacy has been established in 1993, several theories have
been proposed which present similarities to customer intimacy.
This chapter will elaborate on the value discipline customer intimacy
and analyze its distinctive characteristics. Section 2.1 will develop
customer intimacy according to its original definition proposed by
Treacy & Wiersema (1993, 1997), and highlight its differences to two
other value disciplines, namely product leadership and operational
excellence. Section 2.2 will establish why customer intimacy is anchored in the theory of relationship marketing and is related to the
modern perspective on services called the service-dominant logic.
Finally, section 2.3 will outline more specifically three approaches
related to the implementation of customer intimacy, namely key account management, market orientation, and customer relationship
management.

20

2. Towards Customer Intimacy

2.1. Three Value Disciplines to Achieve
Market Leadership
The concept of customer intimacy was first introduced by Treacy &
Wiersema (1993) and emerged from a systematic three-year analysis
of 80 corporations acting worldwide in 40 different business to business (B2B) and business to consumer (B2C) markets. The goal of
this research was to identify patterns among organizations that were
market leaders. Treacy and Wiersema established objective criteria
to explain the reasons for the success or failure of these firms. They
were able to thereby uncover previously hidden sources of competitive advantage for companies operating in these markets. In order to achieve this objective, they analyzed the different facets of
these organizations. First, they considered the value propositions of
these firms, looking at the implicit promises made to their customers.
These value propositions include multiple factors such as price, performance, quality, and scope of the offering. They subsequently analyzed the operating models of the organizations, which include all
the components required to deliver value to the customer, such as
business processes, business structures, management systems, culture, and information technology. Finally, they introduced the novel
concept of value disciplines. Value disciplines are types of strategy
on which the strategy is aligned and are accordingly different from
strategy or strategic goals.
In their study, Treacy and Wiersema identified three distinct value
disciplines: operational excellence, product leadership, and customer
intimacy. As depicted in figure 2.1, operational excellence refers to
a focus on delivering the highest price-quality ratio, or so called best
total cost for the customer. Product leadership concerns organizations that provide their customers with the highest quality and most
advanced innovations. It can be summarized as best product. Finally,
customer intimacy driven organizations may not deliver the cheapest
solution nor the latest innovations to the market, but instead of focusing on average market requirements, they have the ability to develop
individualized solutions, tailored to the exact needs of each of their
customers. This value discipline can be understood as providing

2.1. Three Value Disciplines to Achieve Market Leadership

21

customers with the best total solution. These three value disciplines
shares similarities with the generic competitive strategies proposed
by Porter (2004). The value discipline operational excellence is close
to the generic competitive strategy “overall cost leadership”. Product leadership resembles the strategy “differentiation” and customer
intimacy shows commonalities with the strategy “focus”.

Product Leadership
“Best Product”
Product
Differentiation

Operational
Competence

Customer
Responsiveness

Operational Excellence

Customer Intimacy

“Best Total Cost”

“Best Total Solution”

Figure 2.1.: Three Value Disciplines to Achieve Market Leadership (Treacy & Wiersema, 1997, p.45)
The central argument of Treacy and Wiersema’s thesis is that, in order to become a market leader, an organization should choose one
and only one value discipline and align the two other facets accordingly, that is the value proposition and the operating model of the
organization. The chosen value discipline is the one by which the
organization will gain its market reputation and achieve clarity in
the perception of its customers. It determines all subsequent business decisions related to the value proposition and to the operating
model.

22

2. Towards Customer Intimacy

Selecting one specific value discipline, however, does not mean that
the other two value disciplines become irrelevant for the organization. Treacy and Wiersema nuance their argumentation by stating
that organizations should strive for excellence in one of the three
value disciplines, and achieve a certain threshold in the other two.
Indeed, an organization following operational excellence will not be
successful if its products or services do not achieve a certain degree
of quality. A customer intimacy driven company must keep its prices
within reasonable limits for the customers and deliver high quality
solutions. However, instead of being the first one on the market
to propose a new feature, this company will work closely with its
customers, evaluate to which extent and in which form they need
this feature, and deliver it later than product leadership driven firms
would, but in a way that fits its customers’ requirements. Treacy
and Wiersema argue that organizations striving for excellence in all
three value disciplines do not achieve to become market leaders. As
a matter of fact, this lack of commitment to one value discipline leads
to a dilemma for every business decision taken in the organization.
The trade-off between creating the highest quality product and delivering with the cheapest costs is not performed in a consistent way,
leading to hybrid value propositions which are ambiguous in the
eyes of the customers. In a similar way, the choice to propose standard market offerings rather than to design solutions for individual
customers might be questioned every time a customer raises a new
requirement. This increases complexity significantly, causes confusion in the organization management and leads to “doing business
with yourself rather than with your customers” (Treacy & Wiersema,
1997, p.45).
The next part of this section briefly summarize the value disciplines
operational excellence and product leadership in order to contrast
them in section 2.1.2 with the value discipline customer intimacy.
The core value proposition as well as the operating model are described for each of the three value disciplines as they were originally
presented by Treacy & Wiersema (1997).

2.1. Three Value Disciplines to Achieve Market Leadership

23

2.1.1. Operational Excellence and Product Leadership
as Alternatives to Customer Intimacy
2.1.1.1. The Value Discipline of Operational Excellence
The value proposition of operationally excellent organizations was
labeled by Treacy & Wiersema (1993) as providing the best total cost
to the customer. They consciously used the word cost instead of
price, as the objective is to lower the overall costs incurred to the
customer with the purchased product or service. This includes the
price paid by the customer, but also additional factors such as the
time spent by the customer to purchase the product or to obtain
support. Such companies tend to offer the best price quality ratio on
a limited and precisely defined set of products or services. Low cost
airline companies such as South-West or EasyJet are classic examples
for operational excellence. The promise to the customer is limited to
transporting the customer from departure to destination in a certain
amount of time and very few additional options are available for free
to the customer: the booking and payment must be performed via
internet, there is only one comfort category, and no extra services are
provided during the flight. These enterprises present a clear value
proposition: their customers are aware that they should not expect
anything beyond the standard offering, neither to hope for rewards
for their loyalty, but they also know that the price for these standard
offerings is unmatched by other companies.
Standardization, norms, and procedures are at the heart of the operational excellence operating model. Operational excellence driven
organizations eliminate defects and remove variation in order to lower
costs and to guarantee high quality levels. In this regard, driving
operational excellence shares commonalities with the adoption of
lean management and six sigma programs.1 Operationally excellent
1

Lean management aims to accelerate the velocity of any process by reducing
waste in all its forms. Six sigma is a set of practices originally developed by
Motorola to systematically improve processes by eliminating defects. Six
sigma uses rigorous data analysis to pinpoint the source of errors that
contribute to process variations (George, 2003).

24

2. Towards Customer Intimacy

companies thoroughly standardize their business processes throughout the entire organization and try to include their providers’ activities in these processes. Indeed, vertical integration and tight partnerships with business partners allow them to reduce intermediary
costs such as communication, inventory, and transportation costs.
From a cultural standpoint, the operationally excellent organization
rewards efficiency. Employees are given a set of standard tasks to
accomplish along specific procedures and are expected to complete
them with no variations from the rules. They are rewarded when
they demonstrate a dedication to fulfill the promises made to the
customer, rather than when they show creativity or originality. According to Weill & Ross (2004), firms that pursue operational excellence make large investments in IT systems in which they recognize
the ability to lower costs and, thus, to increase their competitiveness.
They focus on business process management systems2 to centralize
the coordination and control of the activities and to automate routine and non value adding tasks. They also leverage systems that
enable them to automate transactions and facilitate communication
with both customers and providers.3
2.1.1.2. The Value Discipline of Product Leadership
Companies that pursue the product leadership value discipline intend to deliver the best product or service to the market, in terms of
performance and quality, but also with regard to the degree of innovation of their offering. This leads to a clear value proposition, which
target customers who have the willingness to pay a premium fee for
outstanding quality and who value originality and exclusivity of new
features. In many cases, such companies manage to establish an emotional connection to their customers via the provided products and
services. Stern (1997) demonstrated that B2C relationships are more
2

3

Elzinga et al. (1995, p.119) define business process management as a
“systematic, structured approach to analyze, improve, control, and manage
processes with the aim of improving the quality of product and services.
BPM is the method by which an enterprise quality program is carried out.”
See for instance the EDIFACT standard for inter enterprise data exchange
http://www.unece.org/trade/untdid/welcome.htm (accessed on 10.11.2011).

2.1. Three Value Disciplines to Achieve Market Leadership

25

likely to be emotional, while B2B relationships are grounded on a rational basis. Many product leadership driven organizations therefore
predominantly target B2C markets. Apple4 is today’s typical example for product leadership: Apple delivers a clear message, arguing
they provide the highest quality and the most innovative phones and
computers. The market recognizes that they keep this promise and
hails the fact that their products are almost always radically different
from, and significantly better than, those of their competitors. Even
though Apple has also reached certain thresholds in operational excellence and customer intimacy, this company has achieved its reputation by differentiating its products from more standard market
offerings.
In contrast to operationally excellent organizations, which are driven
by procedures and a thorough attention to maintain costs at a low
level, the operating model of product leadership companies emphasizes talents of their employees for generating, promoting, and implementing new ideas. Product leaders rely on their ability to innovate and to bring new forms of value to their customers. They
need to develop management and organization systems that reward
creativity and foster collaboration among the employees. Therefore,
experimentation and risk are key aspects of their culture: employees
are given the time and resources to try new ideas, and to validate
their potential value on the market. Product leaders tend to be organized in small business units having high degrees of autonomy.
This form of structure, described as a federal model by Weill & Ross
(2004), promotes the development of an entrepreneurial and risk
prone environment, as it simplifies and speeds up decision-making
processes.
The core business processes of such organizations are two-fold. On
the one hand, internally, they design business processes that encourage the diffusion of knowledge and expertise to promote new ideas.
These processes support the coordination of the various activities,
4

Apple Inc. reported its best results ever in year 2010 with 71% revenues
growth and 78% earnings growth. Further details are available at
http://www.apple.com/pr/library/2011/01/18results.html (accessed on 17.10.2011).

26

2. Towards Customer Intimacy

but not in the sense operationally excellent organizations perform
it, with strict and rigorous control: they preserve a certain degree of
freedom and autonomy for employees and teams. On the other hand,
focusing on the external perspective, the business model of product
leaders remains sustainable only if the company is able to anticipate
the needs of the market and brings its offering to the market before
competitors do.5 Therefore, the second part of their core processes
seeks fast commercialization and market exploitation. They design
processes that allow them to reduce time to market, by speeding up
engineering, production and delivery phases. Product leaders also
have effective communication campaigns for their new product or
services launch and well established marketing plans. Indeed, their
customers must acknowledge the superior value of their offering in
order to accept the payment of a premium fee. Thus, they need to
be able to clearly articulate the benefits of their value proposition
to prepare the market, and to develop a demand for products and
features that did not exist in the past.

2.1.2. The Value Discipline Customer Intimacy
This section outlines the characteristics of the value discipline customer intimacy, as first introduced by Treacy & Wiersema (1993,
1997), and contrasts them with those of the previously described
value disciplines operational excellence and product leadership. While operationally excellent companies focus on lowering total costs
and product leadership firms try to bring the best products on the
market, customer intimacy driven organizations aim at providing
each of their customers with the best solution.
2.1.2.1. Value Proposition
The uniqueness of customer intimate organizations is that, instead of
focusing on the market and trying to fulfill the most demanded market requirements, they are able to focus specifically on each of their
5

For instance, the Apple iPad arrived on the market in April 2010, a full year
earlier than similar products from competition.

2.1. Three Value Disciplines to Achieve Market Leadership

27

customers and their individual needs, problems, expectations. They
demonstrate to their customers a clear value proposition which goes
beyond mere delivery of products and services: customer intimate
organizations apply their knowledge to investigate the customer’s
specific problems, in cooperation with the customer employees, and
design solutions that include customized versions of the products
and services they intend to sell. Then, they actively control the deployment of the solution in order to ensure that the customer’s expectations are actually reached. Customers see such companies as
trusted partners on which they can rely upon. It may be that other
alternatives on the market are cheaper or more innovative, but a customer intimate organization brings the confidence to its customers
that its solution is solid, tested, and actually delivers the expected
benefits. In fact, customer intimate organizations demonstrate their
commitment to their customers by assuring that their solutions will
deliver the promised results in a mutually beneficial manner: while
customers focus on the part of their operating models that are critical to their own success, the customer intimate firm manages their
secondary processes. For instance, the customer intimate firm will
take over the IT organization of its customers in the form of an
outsourcing contract which guarantees that certain levels of performance, quality and flexibility are achieved.
In order to develop a sustainable competitive advantage, customer
intimacy driven organizations not only fulfill the needs of their customers, but they anticipate the customer problems and identify sources of value for their customers in order to create some demand in
the customer organization. Therefore, companies that pursue a customer intimacy strategy heavily rely on their insight of the customer
industry, on the customer related knowledge they acquired, and on
the interpersonal relationship they developed inside the customer
organization. In that regard, Abraham (2006, p.1) complements the
original definition of customer intimacy and states that customer intimacy is concerned with “the formal or informal set of relationships
established between suppliers and customers, with a diverse array
of partners, from corporate leadership to functional leadership (engineering, marketing, operations, maintenance, or service) and end-

28

2. Towards Customer Intimacy

users of products or services”. While operationally excellent organizations benefit from their optimized processes and product leaders
take advantage of their innovations, customer intimates firms main
asset is the loyalty of their customers (Treacy & Wiersema, 1997).
2.1.2.2. Operating Model
Figure 2.2, originally presented in Treacy & Wiersema (1997, p.130),
depicts the customer intimacy operating model. In order to deliver
tailored solutions to their customers, customer intimate companies
need to establish an operating model that allow them to provide
a broad and deep level of support and services to their customers.
This means that all entities of the enterprise, sales and services, but
also product development and manufacturing must be oriented towards the objective to satisfy the needs of the customer. If a sales
representative thoroughly understands the needs of the customer,
but he is restrained to provide an adequate solution because his
organization does not adapt a product or a service, then this enterprise does not achieve customer intimacy. Treacy & Wiersema
(1997, p.133) call this “customer responsiveness”, carefully understanding the customer needs, showing empathy for the customer
problems, but not being able to provide a satisfactory offering to
the customer. Batt (2004, p.172) confirm that “the firm must keep
deepening its knowledge of the customers and put this knowledge
to work through the organization.” Consequently, customer intimate
companies must empower the front line employees that have understood the customer requirements and provide them with the means
to leverage the skills and capabilities of the entire organization to
build the solution. Such companies, therefore, emphasize structures
aligned with the customer base and the development of decentralized entrepreneurial account teams, who take responsibility for budgets, prices, technological choices, and communication.
Moreover, an organization can rarely deliver a total solution to its
customers solely out of its own assets. The wide variety of customer needs and requirements lead customer intimate companies
to develop partnerships with subcontractors. Treacy & Wiersema

2.1. Three Value Disciplines to Achieve Market Leadership

29

Culture
• Client and field driven
• Variation:“have it your way
mindset“

Organization
• Entrepreneurial client teams
• High skills in the field

Core Processes
• Client acquisition and

development
• Solution development
• Flexible and responsive work
procedures

Management Systems
• Revenue and share of wallet
driven
• Rewards based in part on
client feedback
• Lifetime value of client
analysis

Information Technology
• Customer database linking

internal and external
information
• Knowledge bases built around
expertise

Figure 2.2.: Customer Intimacy Operating Model (Treacy & Wiersema, 1997, p.130)
(1993, p.137) argue that customer intimate organizations are built
upon “hollow delivery systems” and their strength “lies not in what
they own, but in what they know and how they coordinate expertise
to deliver solutions.” Customer intimate organizations tend to take
the role of a resource integrator between the customer and a large
range of operationally excellence and product leadership driven contractors to ensure that the customer receives the best offer, in terms
of features, price, and quality. Consequently, the core processes of
customer intimate organizations should be based on flexible and
solution-driven work procedures that facilitate collaboration inside
the organization as well as with business partners.
Not all customers are responsive to a customer intimacy driven strategy. Several enterprises will take their decision based on price or
product features and, thus, are reluctant to pay a premium fee for the

30

2. Towards Customer Intimacy

value of the expertise provided by customer intimate firms. Organizations which are most inclined to partner with a customer intimate
company exhibit some specific characteristics with regards to their
attitude, operational fit, and financial potential (Treacy & Wiersema,
1997, p.139). The attitude refers to the willingness of the customer to
engage in a business relationship. Indeed a relationship exists only
if the customer perceives a mutual benefit, an opportunity for an
ongoing association, and if he has the readiness to loose some independence in return (Donaldson & O’Toole, 2007, p.58). A customer
that does not demonstrate a certain degree of loyalty should not remain a target of firms pursuing a customer intimacy strategy. The
second characteristic refers to the operational fit. This operational fit
exists if the provider’s expertise matches a deficit of competence on
the customer side. Indeed, if the customer recognizes the superior
skills of the provider, he will rely on him to provide the overall solution. However, if the customer is already too knowledgeable in the
concerned area, he may favor another offering and find it himself on
the market. Since most organizations have developed an expertise
in their core business, customer intimate organizations mainly look
for this expertise gap in the supporting functions of the organization, such as information and communication technology, finance, or
communication in order to create this operational fit. The last characteristic relates to the customer’s financial potential. This financial
potential should be large enough as well as distributed on a longterm period of time for the customer to be a target of the customer
intimate organization. Customer intimate companies invest significant amounts of time and resources to gather and manage customer
related knowledge as well as to generate knowledge that is relevant
for the customer, such as insight in his industry. The return on this
investment is derived from long-term regular cash-flows from the
customer, and short-term single transactions are not profitable for
organizations that base their business model on customer intimacy
(Treacy & Wiersema, 1997, p.140).
As a result, the management system of customer intimate organizations should support the identification and acquisition of customers
presenting such characteristics as well as to help retaining them. Its

2.2. Customer Intimacy: Grounded in Relationships and Services

31

key performance indicators are not related to market shares, but to
account penetration, shares of customers’ spendings, and customer
retention. Similarly, the sales force is driven by two objectives: they
should acquire new customers and they should provide an ongoing
support to their existing customers. The network of interpersonal
relationships established between the employees of the customer intimate firm and the employees of its customers is, therefore, a key
factor of success (Gummesson, 2008, p.91). From a technological perspective, as presented in the introduction, customer intimate organizations develop customer relationship management systems that
allow them to achieve a single view of the customers, as well as
knowledge database to foster the dissemination of customer knowledge.
The next section sets out the similarities between customer intimacy
and the established concepts of relationship marketing, service marketing, and service-dominant logic.

2.2. Customer Intimacy: Grounded in
Relationships and Services
This section establishes the relation between customer intimacy and
the notions of relationship marketing, service marketing, and the
modern view on services, namely the service-dominant logic. Section 2.2.1 introduces the notion of relationship marketing and contrasts it with the transactional perspective on marketing. Then, section 2.2.2 outlines the importance of services for the development of
relationship marketing and elaborates on a relationship marketing
perspective which is particularly important in the scope of this thesis, namely the “Nordic School of Thought” (Gummesson, 1996). In
section 2.2.3, the service-dominant logic is presented as the evolution of relationship marketing. Finally, in order to fully understand
the concept of customer intimacy, section 2.2.4 contrasts customer
intimacy with relationship marketing and with the service-dominant
logic.

32

2. Towards Customer Intimacy

2.2.1. Two Divergent Perspectives on Marketing
While the notion of marketing as a distinct discipline arose in the
beginning of the 20th century, the emphasis on relationships has only
received attention over the past 40 years (Sheth & Parvatiyar, 2000).
Arndt (1979), introducing the concept of “domesticated markets”,
was one of the first to establish the importance of developing longlasting relationships with key customers rather than focusing on single transactions. Adler (1966) and later Varadarajan & Rajaratnam
(1986) outlined the widespread acceptance of symbiotic marketing
as a way to achieve sales and profit growth, with an emphasis on
collaboration and strategic partnership for mutual benefit of the parties. The first formalization of the concept of relationship marketing
in literature occurred in 1983: Berry (1983, p.25) defined relationship
marketing as “attracting, maintaining and – in multi-service organizations – enhancing customer relationships.” Since then, multiple
perspectives have emerged with an emphasis on a variety of themes
such as quality, customer service, alliance and partnerships, communication and interaction (Mohr & Nevin, 1990; Christopher et al.,
1993; Morgan & Hunt, 1994; Varadarajan & Cunningham, 1995).
The first part of this section focuses on the exchange perspective and
defines transactional marketing. Then, the second part elaborates on
the relationship perspective and defines relationship marketing in
contrast to transactional marketing.
2.2.1.1. Transactional Marketing
The concept of transactional marketing originated in the industrial
era, as a consequence of mass production, mass consumption, and
the division of labor. In these provider driven markets, the most important challenge was to optimize production capabilities and employees productivity in order to increase the produced volumes, thereby
achieving economies of scale. Low priced and standardized products
replaced customized offerings. New specialized organization structures with dedicated purchasing and selling functions fundamentally
changed the way of doing business. Sheth & Parvatiyar (2000) argue

2.2. Customer Intimacy: Grounded in Relationships and Services

33

that the providers and customers have been separated. Business relationships have become impersonal or even replaced by intermediaries,6 such as wholesales companies and distributors, whose roles,
acting as agents, are two-fold: first, they have to handle the stocks
produced and, second, they have to distribute these goods into the
market.
The exchange perspective on marketing arose in the early 1960s when
most markets became saturated and competition increased (Wengler,
2006). New marketing practices emerged, “focused on sales, advertising and promotion, for the purpose of creating new demand to
absorb the oversupply of goods” (Sheth & Parvatiyar, 2000, p.130).
Marketing functions were implemented as a means to locate and
persuade potential customers to purchase more goods and services
in order to increase sales volumes and generate additional profits
(Gruen, 1997). In this perspective, customers are not considered as
single and active entities, but aggregated in passive market segments.
As depicted in figure 2.3, the focus is on the outcome of the transactions: marketing aims solely at supporting sales activities, rather
than on developing and maintaining business relationships. Value
creation and value distribution are two distinct activities, and marketing concentrates on the latter one only: the customer is solely considered as the receiver of value distributed by the firms, and does not
participate in the creation of value.
2.2.1.2. Relationship Marketing
In the 1980s, as customers expectations were raising, companies started
to search for new means of generating value. The exchange perspective focused on single transactions was questioned and new programs dedicated to the partnership with individual customers emerged (Shapiro & Wyman, 1981). Later, several publications recognized
the potential of collaboration and cooperation between buyers and
sellers to develop a competitive advantage (Narver & Slater, 1990;
Varadarajan & Rajaratnam, 1986). The findings from Reichheld &
6

These intermediaries are called “middlemen” in Sheth & Parvatiyar (2000).

34

2. Towards Customer Intimacy

Process

Relationship
Perspective

Value Creation

Value Distribution

Exchange
Perspective
Outcome

Figure 2.3.: Exchange and Relationship Perspectives (Sheth & Parvatiyar, 2000)

Sasser (1990) that a 5% improvement in customer retention can result in a profitability improvement comprised between 25% and 85%
created a strong impulse for research that investigates the association
between customer loyalty, retention, and satisfaction (Dick & Basu,
1994). Later, various studies argued that customer satisfaction could
be better achieved through an emphasis on customer relationships,
with the objective to retain valuable customers, rather than through
a focus on single transactions (Day & Montgomery, 1999). Moreover,
the development of new technological solutions and the growth of
the service economy changed the market dynamics and boosted the
development of the relationship marketing concept. Indeed, new IT
based solutions and the rise of internet services enable selling and
buying firms to reestablish direct contact, without the needs of in-

2.2. Customer Intimacy: Grounded in Relationships and Services

35

termediaries.7 The growth of the service economy and the ongoing
shift to servitized businesses further reduce the needs for intermediates, as services are often directly provided by the provider. Indeed,
as developed in section 2.2.2, services enable the development of relationships and much literature on service marketing is devoted to
relationship marketing (Grönroos, 2007; Lovelock & Wirtz, 2007).
As illustrated in figure 2.3, Sheth & Parvatiyar (2000) defined the
two axes of the relationship perspective in contrast to the exchange
perspective. The first dimension outlines that value is not only distributed to the customer, but created in collaboration with the customer. Higher value can be generated if the customer actively cooperates with the provider and shares his knowledge and expertise.
Marketing, thus, should not unilaterally focus on convincing customers of the benefits of the value proposition, but it should involve
the customers in the definition and development of a joint value
proposition which is mutually beneficial for both parties. The second
dimension refers to the process dimension of relationship marketing. Relationship marketing requires to establish a set of processes
focused on the initiation, maintenance, and termination of business
relationships, rather than on the outcomes on the relationship. In line
with these two dimensions, Parvatiyar & Sheth (2000, p.9) propose
to define relationship marketing as “the ongoing process of engaging
in cooperative and collaborative activities and programs with immediate and end-user customers to create or enhance mutual economic
value, at reduced cost.”
The next part of this section emphasizes the importance of services
for the development of relationship marketing and associates service
marketing to relationship marketing.

2.2.2. The Service Dimension of Relationship
Marketing
Services in contemporary times have become the most important
driver of Western economies as they represent over 70% of the Gross
7

Parvatiyar & Sheth (2000) call this transformation the deintermediation
process by which producers and customers directly interact with each other.

36

2. Towards Customer Intimacy

Domestic Product (GDP) in both Europe8 and United States.9 However, it remains challenging to characterize services as they refer to
a wide variety of concepts, and were described differently in various disciplines such as information technology, service design, or
marketing. Focusing on business services and on the marketing perspective, Grönroos (2007, p.25) states that a service is “a process
consisting of a series of more or less intangible activities that normally, but not necessarily always, take place in interactions between
the customer and service employees and/or physical resources or
goods and/or systems of the service provider, which are provided
as solutions to customer problems.” In this definition, three aspects
characterizing services are important: customer interaction, service
intangibility, and service individualization. Bruhn & Georgi (2006)
confirm the relevance of these characteristics in their three dimensional continuum along which both products and services are positioned: the three dimensions of this continuum are interactivity,
intangibility, and individuality. They argue the more an offering is
interactive, intangible, and individualized, the more this offering is
considered as a service. A detailed analysis of these characteristics
establishes the reasons why relationships are embedded in services
and why service marketing closely relates to relationship marketing:
• Customer interaction refers to the involvement of the customer
in the process of delivering the service. Several activities of
this process include the customer employees and provide them
with the opportunity to communicate, exchange information
and knowledge and, thus, to participate to some extent to the
creation of value, together with the provider employees. The
value of a service is not consumed as an outcome by the customer at the end of the service process as a product would be,
but simultaneously during the service process. Consequently,
services are by definition aligned to the two dimensions of the
previously presented relationship perspective: a focus on the
8
9

See https://www.cia.gov/library/publications/the-world-factbook/geos/ee.html
(accessed on 10.11.2011).
See https://www.cia.gov/library/publications/the-world-factbook/geos/us.html
(accessed on 10.11.2011).

2.2. Customer Intimacy: Grounded in Relationships and Services

37

value creation process rather than on the value creation outcome, and a value which is created with the customer rather
than distributed to the customer.
• Intangibility refers to the fact that a service do not systematically result in a tangible outcome. Travel or hotel services for
instance are intangible: customers who purchase a service experience are left without any “tangible good” at the end of the
service delivery. More specifically, customers cannot see and
test the service characteristics prior to purchasing it, as they
would with a product. They have to trust the provider in its
ability to deliver the agreed service, and to demonstrate a willingness to establish business relationships with their trusted
service partners. This fact is confirmed by several studies which
acknowledge the positive association between trust and relationship in a service context (Palmatier et al., 2006; Berry, 1995;
Morgan & Hunt, 1994).
• Individualization concerns the ability of the provider to customize
his offering in order to fulfill the customer requirements. This
aspect refers to the strategic perspective on relationship marketing. Indeed, the concept of relationship marketing emerged
as companies were seeking new sources of competitive advantage and new means to generate value. Considering each customer on an individual base rather than focusing on market
segments is the essence of relationship marketing and, therefore, the servitization of the offering is the means to adopt a
relationship marketing strategy (Berry, 1983, p.26).10
The “Nordic School of Thought”, originating in Sweden and Finland,
is recognized as the pioneer and leader in service marketing. It has
established a direct association between service marketing and relationship marketing (Gummesson, 1996). This approach is led by two
10

Berry (1983) establishes the five strategy elements for practicing relationship
marketing: “developing a core service around which to build a customer
relationship, customizing the relationship to the individual customers,
augmenting the core services with extra benefits, [...] pricing services to
encourage customer loyalty, and [...] marketing to employees so that they, in
turn, will perform well for customers.”

38

2. Towards Customer Intimacy

prominent researchers in the field of relationship marketing, Grönroos and Gummesson, who elaborated two definitions of relationship
marketing:
• Grönroos (1997, p.407) presents a definition which is close to
the original one proposed by Berry (1983), and emphasizes the
notion of relationship process and the development of a partnership to achieve the objectives of both parties. He states
that relationship marketing is “the process of identifying and
establishing, maintaining, enhancing, and when necessary terminating relationships with customers and other stakeholders,
at a profit, so that the objectives of all parties involved are
met, where this is done by a mutual giving and fulfillment of
promises.”
• Gummesson (1995)11 proposes a definition that emphasizes the
notion of interactions. He argues that relationship marketing is
“marketing seen as interactions, relationships, and networks,”
where “relationships are contacts between two or more people, but they also exist between people and objects, symbols
and organizations”. He also defines networks as “sets of relationships”, and interactions as “activities performed within
relationships and networks” (Gummesson, 1996, p.33). This
perspective is particularly relevant for this contribution, as the
model proposed in chapter 5 is based on an analysis of interaction data.
The next part of this section introduces the service-dominant logic as
an evolution of the concept of relationship marketing.

2.2.3. The Service-Dominant Logic as an Evolution of
Relationship Marketing
In 2004, the prestigious Journal of Marketing published an article entitled “Evolving to a new dominant logic for marketing” (Vargo &
Lusch, 2004a) which brings the concepts of relationship marketing
11

This citation is presented in Gummesson (1996).

2.2. Customer Intimacy: Grounded in Relationships and Services

39

and service even closer. This article includes controversial ideas that
were discussed at length before and after its publication: it was accepted for publication only after a five year review process (Bolton,
2006). The authors claim that “marketing has moved from a goodsdominant view, in which tangible output and discrete transactions
were central, to a service-dominant view, in which intangibility, exchange processes, and relationships are central” (Vargo & Lusch,
2004a, p.2). They also bring accordingly a new perspective on the
service concept which they define as “the application of specialized competences (knowledge and skills) through deeds, processes,
and performances for the benefit of another entity or the entity itself” (Vargo & Lusch, 2004a, p.2).
The authors’ argumentation for a new logic of marketing consists
of multiple foundational premises which are thoroughly described
with reference to past literature mainly rooted in the field of relationship marketing (Vargo & Lusch, 2008a, p.7). The eight original
foundational premises are italicized in this and the next paragraphs.
First, in contrast to many theories that differentiate tangible goods
on one side and services on the other side, the thesis of Vargo and
Lusch unifies goods and services: goods exchanged within economic
transactions are actually the “outcome” of services understood as application of the provider’s knowledge and skills. Thus, even though
goods are traded between economic actors, “service is the fundamental basis of exchange.” Instead of separating goods from services, the
service-dominant logic distinguishes two types of resources: the operant and the operand resources. Operant resources are active and
possess knowledge and skills, which have to be applied on other resources in order to generate value. For instance, if someone has some
special knowledge or skills, but does not utilize them, then no value
is created. On the other hand, the operand resources, such as goods
and natural resources, are static and inherently contain the outcomes
of the application of the operant resources knowledge and skills: the
operant resources perform transformation actions on the operand resources in order to increase their potential value, once the customer
utilize them. Therefore, goods, in this logic, embed the knowledge
and skills of the provider: “goods are distribution mechanisms for service

40

2. Towards Customer Intimacy

provision”.
As described in section 2.2.1.1, the emergence of intermediaries in
the 20th century, such as distributors and wholesales companies, has
led to separate providers and customers who are no longer in direct contact: providers trade with intermediaries and intermediaries
trade with customers. It has resulted, on the long term, in hiding
that the application of knowledge and skills is the essence of economic transactions: “indirect exchange masks the fundamental basis of
exchange.” Moreover, Vargo and Lusch argue that the dominance of
the manufacturing perspective and the segmentation of the economy
into eras, such as the agricultural and later the industrial era, have
focused the analysis of economic activities on the optimization of
goods production efficiency. As a result, the tangibility dimension
has received an overly important consideration and intangible items
have been simply perceived as side elements. In contrast to this perspective, the service-dominant logic states that intangibility is only
one aspect that characterizes economic exchanges and, therefore, “all
economies are service economies.” Since all economic exchanges are derived from the application of operant resources such as knowledge
and skills, the service-dominant logic states that “operant resources
are the fundamental source of competitive advantage.” Indeed, the added
value that leads to a competitive advantage may lie in the characteristics of the product or service sold to the customer, but primarily
results from the leverage of knowledge and skills to fulfill the customer’s needs and solve his problem.
The remaining characteristics of the service-dominant logic outline
its association with relationship marketing. Vargo and Lush argue
that value, as for relationship marketing, is created together with the
customer rather than distributed to the customer: “the customer is always a co-creator of value.” They emphasize the role of the customer as
an operant resource that uses its skills and knowledge as well as the
significance of the interaction with the customer in order to increase
the created value. Then, the service-dominant logic also differentiates the value-in-exchange from the value-in-use and “the enterprise
cannot deliver value, but only offer value propositions.” Indeed, even
in the case of manufactured goods, the actual value is only created

2.2. Customer Intimacy: Grounded in Relationships and Services

41

once the customer is using the good. As long as this is not the case,
the provider has only proposed value to the customer. Finally, the
authors state that “a service-centered view is customer oriented and relational,” and describe in four arguments this service-centered view:
“(1) identify or develop core competences, the fundamental knowledge and skills of an economic entity that represent potential competitive advantage; (2) identify other entities (potential customers) that
could benefit from these competences; (3) cultivate relationships that
involve the customers in developing customized, competitively compelling value propositions to meet specific needs; (4) gauge market
place feedback by analyzing financial performance from exchange to
learn how to improve the firm’s offering to customers and improve
firm performance” (Vargo & Lusch, 2004a, p.5).
In summary, the service-dominant logic is closely aligned with the
relationship perspective (Vargo & Lusch, 2006) and shares its two dimensions “value creation” and “process”.12 First, in both approaches
the value is co-created by the provider and the customer rather than
created by the provider and distributed to the customer: the customer is an active participant, an operant resource. Second, the focus
is on the activities that lead to value creation rather than on the outcome. These activities, called the value creation process in relationship marketing are defined in Vargo & Lusch (2004a, p.2) as “the application of specialized competences (knowledge and skills) through
deeds, processes and performances” and represent the substance of
economic exchanges in the service-dominant logic. Importantly, the
service-dominant logic strengthen the significance of knowledge and
its application in order to individualize the value proposition and
achieve a competitive advantage even more than relationship marketing: knowledge has already established as an important dimension
in relationship marketing, but it is, in the service-dominant logic,
12

The authors, however, do not oppose the service-dominant logic to the
exchange perspective. As they compare the service-dominant logic with the
exchange perspective, Vargo & Lusch (2006, p.48) consider that the
service-dominant logic “bridges the exchange and relationship perspective
and, therefore, obviates the apparent need for abandoning the exchange
paradigm.”

42

2. Towards Customer Intimacy

defined as the fundamental source of competitive advantage.
The next part of this chapter establishes why customer intimacy is a
value discipline grounded in the concepts of relationship marketing
and service-dominant logic.

2.2.4. Customer Intimacy: A Relationship and Service
Based Value Discipline
It is possible to establish an association between customer intimacy,
relationship marketing, and the service-dominant logic by comparing the previously introduced descriptions of these three notions. In
contrast to the value disciplines product leadership and operational
excellence, customer intimacy is rooted in the concept of relationship
marketing and it shares commonalities with the service-dominant
logic. This statement is motivated by the following four arguments
which are elaborated in the next paragraphs: similarly to relationship marketing and the service-dominant logic, (i) customer intimacy
supports the idea that value is co-created by the provider and the
customer; (ii) customer intimacy focuses on relationship processes
established with customers rather than on the delivery of produced
outcomes; (iii) customer intimacy does not specifically distinguish
tangible products and intangible services; and (iv) customer intimacy recognizes that knowledge is the main source of competitive
advantage.
• The provider and the customer co-create value
The first dimension of the relationship perspective which is described in section 2.2.1.2 states that value is created together
with each customer rather than produced by the provider and
distributed to customers. Customer intimacy, with its focus on
individual customers needs, is closely aligned to this perspective. Customer intimate organizations do not propose solutions
fitting most demanded market requirements, but closely cooperate with the customer in order to understand his needs
and requirements, thereby providing a perfectly suited solution. Quoting executive management from a customer intimacy

2.2. Customer Intimacy: Grounded in Relationships and Services

43

driven organization, (Treacy & Wiersema, 1997, p.41) state that
“the product is conceived at the customer’s office”. Moreover,
the customer intimate organization not only provides the solution, but also ensures that, once deployed, the solution fulfills
the customer expectations: they take responsibility for results.
Therefore, the focus of customer intimacy is not on the value in
exchange but on the value in use, as perceived by the customer.
• The emphasis is on the process rather than on the outcome
The second dimension of the relationship perspective emphasizes the notion of process, consisting in multiple interactions
with the customer on a long-term perspective, instead of considering the outcome of single transactions in the short-term.
This view is shared by customer intimacy driven companies,
for which most relevant key performance indicators are based
on long-term customer lifetime value and customer retention
rates rather than on market shares at a specific point in time.
Indeed, customer intimacy firms invest in the customer in the
initial interactions in order to grow inside the customer organization and to leverage the existing potential in its operations.
Therefore, the operating model of customer intimacy driven organizations is built around the relationship process established
with the customer.
• The offering can be tangible or intangible
The third similarity refers to the absence of distinction based
on the degree of intangibility of the value proposition in the
definition of the customer intimacy. The focus is on the solution, which consists of a combination of all required elements
to fulfill the customer’s needs. These elements can be tangible
goods as well as intangible services. In this sense, customer intimacy is close to the service-dominant logic perspective which
considers that the tangibility dimension is not the most important factor: “all economies are service economies” (Vargo &
Lusch, 2004a, p.10). Moreover, as in relationship marketing,
customer intimacy insists on the previously presented the degree of individualization: customer intimate firms rely on their

44

2. Towards Customer Intimacy

ability to customize and individualize their offering to the customer in order to achieve a competitive advantage.
• Knowledge is the main source of competitive advantage
Finally, the emphasis on knowledge is the fourth commonality between these concepts. While the service-dominant logic
states that knowledge in a broad sense is the fundamental source
of competitive advantage, customer related knowledge, insight
in the customer business and the ability to use them are key
differentiators of customer intimacy and relationship marketing. Indeed, close similarities can be found among the following two statements. Focusing on customer intimacy, Treacy
& Wiersema (1997, p. 131) argue that “deep customer knowledge and breakthrough insights about the client’s underlying
processes are the backbone of every customer-intimate organization.” Focusing on relationship marketing, Grönroos (2007,
p.30) considers that a “key requirement in relationship marketing strategy is that a manufacturer, wholesale, retailer, service firm, or supplier knows the long-term processing needs
and desires of their customer better and offers value on top of
the technical solutions embedded in consumer goods, industrial equipment or services.” In addition, Berry (1995, p.153)
confirms that “relationship marketing allows service providers
to become more knowledgeable about the customer’s requirements and needs.” It is established, thus, that customer intimacy, relationship marketing, and the service-dominant logic
all recognize the significance of knowledge and specifically customer related knowledge.
In conclusion, this analysis demonstrates that customer intimacy is a
type of strategy which is strongly related to services, closely aligned
with relationship marketing, and which shares multiple similarities
with the service-dominant logic. In the next section, this thesis describes three approaches related to the adoption of the customer intimacy value discipline.

2.3. Three Approaches Related to Customer Intimacy

45

2.3. Three Approaches Related to Customer
Intimacy
This thesis aims at providing a model and a methodology for assessing and monitoring customer intimacy in B2B markets and, therefore,
to support the relationship marketing activities of B2B providers. In
that sense, this contribution relates to existing approaches for adopting a marketing strategy. The objective of this section is to elaborate
on the similarities between customer intimacy and three marketing
approaches, namely key account management, market orientation,
and customer relationship management. While key account management focuses on individual relationships with the most important customers in a B2B context, market orientation defines a culture
centered around the management of customers’ and competitors’ related knowledge, and customer relationship management allows the
organization to focus on most profitable business relationships.

2.3.1. Key Account Management
The concept of key account management has emerged over the last
40 years along with the development of relationship marketing. In
the literature, it was also referred to as large account management,
global account management, or strategic account management (Holt
& McDonald, 2000; Boles et al., 1999). The rationale of key account
management is to develop a specific marketing program for the provider’s most important customers in the context of B2B markets. If
a limited number of customers generate the most important share
of revenues and profits, it is sound to allocate dedicated employees
and teams to focus exclusively on the management of the relationships with these customers. According to Cannon & Narayandas
(2000, p.408), key account management is the “embodiment and implementation of the relationship marketing paradigm for large business customers.” Wengler (2006, p.27) defines key account management as “a supplier’s relationship marketing program which aims at
establishing, developing and maintaining a successful and mutually
beneficial business relationship with the company’s most important
customers.”

46

2. Towards Customer Intimacy

From the perspective of the provider, the main objectives of key account management are to ensure customer retention and to maximize the customer value (Wengler, 2006; Cannon & Narayandas,
2000; Berger et al., 2002).
Customer retention means to keep the customer and to ensure that he
will generate regular incomes over time, for instance by purchasing
products or services every quarter or every year. Customer retention has, thus, been established as an indicator of the loyalty of the
customer to the provider (Lam et al., 2004). The motivation for focusing on customer retention builds upon empirical evidence which
demonstrates that it is cheaper to keep a customer rather than to
acquire a new one. Reichheld & Sasser (1990) established that a 5%
increase of the customer retention rate can generate up to 85% improvement in profitability. In order to retain customers, two means
have proven to be successful. Providers can either improve customer
satisfaction or increase switching costs:13 “both enhancing customer
satisfaction and increasing switching costs can be seen as important
strategies that promote customer loyalty” (Lam et al., 2004, p.308).
Consequently, the key account manager responsibilities can be derived from the objective of retaining customers: he should ensure
customer satisfaction by providing solutions that fulfill the customer
needs and expectations, as well as try to increase the switching costs
by making the customer more dependent on the provider capabilities, skills and knowledge.
The second objective of key account management, maximizing customer value, is derived from various analysis establishing that customer retention is not a sufficient condition for being successful: the
business relationships must be profitable (Reinartz & Kumar, 2000).
The important resources committed by the provider, with specific
teams focusing on individual customers, have to lead to a positive return on investment in the long run. Therefore, a significant contribution of key account management to relationship marketing literature
lies in the definition and assessment of different indicators to assess
13

Switching costs are the costs incurred to the customer when changing the
supplier (Lam et al., 2004).

2.3. Three Approaches Related to Customer Intimacy

47

this degree of “long-term profitability” value, such as customer lifetime value, customer equity, and return on relationships. Customer
lifetime value is a monetary approach of the overall value returned
by the customer to the provider. In this perspective the customer is
seen as any other investment of the provider, and the customer lifetime value is calculated as the net present value of the contribution
margin over the relationship lifetime (Berger et al., 2002).14 Customer
equity enlarges this measurement and aggregates customer lifetime
value over all actual and potential customers of the provider in the
industry.15 Finally, return on relationships is estimated from a network perspective and measures the net financial outcome of the overall relationship network of the provider.16 Consequently, in order to
maximize customer value, key account managers are responsible for
minimizing the costs of the relationship, for instance by reducing the
process and transaction costs and by removing uncertainty to make
business relationships more predictable. They are also responsible
for expanding the provider’s business activities inside the account,
by identifying new opportunities for partnership and synergies with
the customer (McDonald et al., 1997).
A characteristic of marketing in B2B markets is that it involves many
individuals from the provider and the customer organizations.17 People with diverse functions, knowledge and skills on both sides participate in the relationship process. For instance, sales employees
actively communicate with the purchasing department and the head
of the customer organization. Services employees cooperate with
various customer employees in order to perform their tasks. Therefore, multiple interactions occur within the scope of the relationship
14
15

16

17

This is calculated as the sum of the discounted earnings (revenues minus
costs) over the lifetime of the relationship (Berger et al., 2002).
Rust et al. (2004, p.110) define customer equity as “the total of the discounted
lifetime values summed over all of the firm’s current and potential
customers.”
Gummesson (2008, p.257) defines return on relationships as “the long-term
net financial outcome caused by the establishment and maintenance of an
organization’s network of relationships.”
“The many-headed customer and the many-headed supplier” is the 6th
element out of the 30 Rs of relationship marketing (Gummesson, 2008, p.91).

48

2. Towards Customer Intimacy

and a network formed by provider and customer employees has to
be coordinated. This coordination task is an essential aspect of the
key account manager’s activities (Holt & McDonald, 2000). Acting
at the interface between both companies, the key account manager
represents the customer inside the provider organization and embed
the customer as far as possible in the provider’s own processes. On
the other side – inside the customer organization – the key account
manager coordinates the provider resources, optimizes their utilization, and ensures that a clear communication is established between
provider and customer employees.
This notion of interaction based relationship network is foundational
for the contribution of this thesis. Chapter 5 introduces the CI Analytics model to infer this relationship network by applying machine
learning algorithms on customer related data. This model is complementarity to key account management: the solution proposed by this
thesis and prototypically implemented in the software CI Analytics18
supports key account managers with regard to their investments decisions and help them coordinate this relationship network.

2.3.2. Market Orientation
The concept of market orientation has originally been proposed in
order to elaborate the actual steps required to implement the marketing strategy, instead of considering marketing as a “business philosophy” (Deng & Dart, 1994).19 Indeed, the focus of market orientation is on specifying a set of activities that a firm should perform to
achieve its marketing objectives, rather than on defining the concept
of marketing itself. Market orientation modifies the firm behavior
with regard to its customers and competitors, and also influences its
organizational structure. This notion has emerged in marketing literature as several studies proved the positive impact of adhering to
18
19

This software is described in chapter 6.
Deng & Dart (1994, p.726) define the marketing concept as a business
philosophy holding that “long term profitability is best achieved by focusing
the coordinated activities of the organization toward satisfying the needs of a
particular market segment.”

2.3. Three Approaches Related to Customer Intimacy

49

market orientation on business performance (Narver & Slater, 1990;
Kohli & Jaworski, 1990; Rodrigez Cano et al., 2004).
Even though this concept was defined in multiple ways, Jaworski
& Kohli (1993, p.54) introduced a definition of market orientation
which is recognized as a reference and which consists of the three
following aspects: “(i) organization-wide generation of market intelligence pertaining to current and future customer needs; (ii) dissemination of the intelligence across departments; (iii) the organizationwide responsiveness to it.” The first aspect - generation of market
intelligence – refers to the ability of the organization to acquire three
different categories of knowledge: knowledge about the customer
needs and preferences, knowledge about competitors and their ability to fulfill these needs, and finally knowledge about the customer
market and environment, which might influence the customer behavior, such as government regulations. The second aspect – intelligence dissemination – refers to the ability of the entire organization
to share this acquired knowledge in a way that reaches the employees
who can use it. In order to achieve this, the firm has to establish both
vertical and horizontal communication structures so that all departments, teams, and employees can easily exchange relevant market
intelligence information. The third aspect – responsiveness – refers
to the action taken in response to the acquired market intelligence.
Gathering and exchanging market intelligence information does not
improve the created value, the competitiveness, or the business performance unless this knowledge is actually leveraged. In order to
react on this knowledge, the firm can, for instance, choose to focus
on specific market segments. It can also promote its offering in a way
that create some interest in the customer organization, or adapt its
products and services to anticipate the customer needs.
According to Narver & Slater (1990) and Deng & Dart (1994), the
firm has to focus on four main dimensions in order to achieve market orientation: customer orientation, competitor orientation, interfunctional coordination, and profit emphasis. Customer orientation
represents the extent to which the firm adopt behaviors demonstrating its commitment to its customers. It refers to the ability of the
firm to obtain and understand its customers needs and to provide an

50

2. Towards Customer Intimacy

adequate response ensuring the satisfaction of its customers. Competitor orientation represents the firm’s ability to gather information
about its competitors and to act upon it. For instance, the firm can
enhance its products or services with new features in order to improve the competitiveness of its value proposition or it can modify
its pricing model. Inter-functional coordination relates to the ability of the different teams and departments of the firm to collaborate,
share information, and coordinate their activities in response to the
acquired customer and competitor intelligence. Finally, profit emphasis reflects the ability of the firm to consider profitability as a key
performance indicator.
The comparison of customer intimacy and market orientation allows
to establish some similarities as well as some differences between
these two concepts. Market orientation is both a broader and narrower concept than customer intimacy. The main similarity consists
of the importance of knowledge and, more specifically, the emphasis
on customer related knowledge in both approaches. While market
orientation insists on gathering market intelligence and acting upon
this information accordingly, customer intimacy focuses on obtaining
knowledge about the customer’s needs and expectations in order to
tailor and shape the offering. Tuominen et al. (2004) confirm this commonality as they established a strong association between customer
intimacy and market orientation. Moreover, both concepts emphasize the need to involve the entire organization, and not only the
marketing department in the process of managing customer related
knowledge: market orientation requires a strong ability to disseminate market intelligence and well established “inter-functional coordination”. Similarly, the customer intimacy operating model requires
that all entities of the organization are focused on solving customers’
problems and empowers the employees in contact with the customer.
However, market orientation is different from customer intimacy as it
does not focus on the customer only, but on the overall market intelligence and includes also knowledge related to the firm’s competitors.
The objective of market orientation is not to fulfill to the highest
extent the needs and expectations of individual customers, as customer intimacy does, but to understand these needs, to understand

2.3. Three Approaches Related to Customer Intimacy

51

the competitive offers available on the market, and to provide a solution which is better than those of competitors. In addition, in market
orientation, the emphasis is solely on knowledge and acting upon
this knowledge: it does not consider the relationship established between the customer and the provider. As opposed to customer intimacy, market orientation is not grounded in relationship marketing:
the objective is not to involve the customer as a partner to co-create
the value. In market orientation, the provider gathers market intelligence and act upon it in order to improve its value proposition for a
specific market segments, but customers do not participate directly
to the design of this value proposition on an individual basis. Moreover, even though some articles related to market orientation refers
to its long-term focus, this is to outline the long-term sustainability
of the firm, rather than the development of long-term relationships
with customers (Narver & Slater, 1990). The focus of market orientation remains the outcomes produced by the firms for its customers
rather than the processes of value creation with its customers.
This comparison of the concepts of market orientation and customer
intimacy leads to the conclusion that customer intimacy cannot be
assessed in the same way as market orientation is measured. Indeed,
while some of these aspects related to the customer related knowledge can be taken into consideration for the evaluation of customer
intimacy, the assessment of customer intimacy must include the customer relationship dimension.

2.3.3. Customer Relationship Management
The concept of customer relationship management (CRM) has become popular in the late 1990s, mainly through its association with
IT, and more specifically with the development of IT based CRM systems, which aim at supporting the management of the relationships
with customers and their underlying interactions. Several software
providers and consulting firms have included CRM in their portfolio, and this market represents in 2010 over $10B (Maoz et al., 2010).
However, CRM cannot be reduced to this technological perspective
without the risk to jeopardize the CRM initiative. Indeed, the fact

52

2. Towards Customer Intimacy

that firms perceive CRM only as a technological project is seen as a
significant reason for the failure of CRM adoption (Doherty & Lockett, 2004). In order to successfully achieve CRM, a change in the
mindset of the organization is required. Hasan (2003, p.16) argues
that to adopt CRM, “companies must make a fundamental change
in the way they do business, modifying their approach to sharing
information and coordinating activities within the company.”
From the technological point of view of software vendors to the
philosophical approach of CRM, considering it as a “business mindset”, the CRM concept has been investigated in numerous ways. In a
thorough literature review, Zablah et al. (2004) identified five dominant CRM perspectives: strategy, process, philosophy, capability, and
technology. This section focuses on the strategic and operational –
process based – perspectives in order to outline the commonalities of
CRM with relationship marketing and customer intimacy.
From a strategic perspective, Payne & Frow (2005, p.168) define CRM
as “a strategic approach that is concerned with creating improved
shareholder value through the development of appropriate relationships with key customers and customer segments.” A similarity can
be perceived between this definition and the two dimensions of the
previously described relationship perspective: both this definition
and the relationship perspective emphasize the process of developing
customer relationships as well as the creation of value for all shareholders, including both the provider and the customer. Parvatiyar &
Sheth (2001) recognize that the terms CRM and relationship marketing have been often used to describe the same phenomenon. More
precisely, the association between relationship marketing and CRM
is in somehow similar to the association between market orientation
and marketing: while market orientation is defined as the implementation of the marketing concept, CRM is described in literature as the
means to adopt relationship marketing. Zablah et al. (2004, p.480)
confirm that “relationship marketing is often cited as the philosophical basis of customer relationship management”. Gummesson (2008,
p.7) further insists on the practical aspects of CRM and defines it
as “the values and strategies of relationship marketing [...] turned

2.3. Three Approaches Related to Customer Intimacy

53

into practical application and dependent on both human action and
information technology.”
From the operational perspective, much literature has focused on
defining CRM as a set of processes. Reinartz et al. (2004, p.294) define CRM as “a systematic process to manage customer relationship
initiation, maintenance, and termination across all customer contact
points in order to maximize the value of the relationship portfolio.”
This is a broad perspective which covers the life cycle of the relationship and which is closely aligned to the definition of relationship marketing presented in section 2.2.1.2. Bueren et al. (2004) and
Gebert et al. (2003) further detail this process perspective and argue
that CRM consists of six sub-processes:
• Campaign management refers to the segmentation of the market
in smaller groups of customer and prospective customers, and
then, to the planning and realization of customized communications and interactions with these targeted groups of customers.
• Lead management refers to the systematic identification and prioritization of potential sales opportunities which raise customers’
interest.
• Offer management, as the core sales activities, relates to the process of qualifying leads with the customer and transforming
them into offers that the customer can purchase.
• Contract management is the process of maintaining and adjusting
long-term contract in order to ensure that customers’ expectations remain fulfilled, even in the case that customers’ needs
have changed.
• Complaint management ensures that all issues encountered by
customers as well as all sources of dissatisfaction are actually
tracked and managed consistently.
• Service management focuses on the maintenance, repair, and support activities related to the customers’ purchases.

54

2. Towards Customer Intimacy

These descriptions of the strategic and operational perspectives on
CRM outline the close association between this concept and relationship marketing. They also highlight an important characteristic
of CRM which distinguishes it from customer intimacy. In contrast
to customer intimacy, CRM does not focus on customizing the value
proposition and adapting the offering in order to fit exactly the needs
of each customer. The goal is not to establish close and collaborative relationships that tend to transform in partnership with all customers. On the contrary, the multiple CRM definitions insist on the
objective to create value for the shareholders and to maximize it by
targeting the most profitable customers. In that sense, CRM does not
exclude transactional relationships as long as they remain profitable.
Such relationships may not generate as much revenue as closer ones,
but they also require a smaller investment in time and resources and,
thus, may be profitable. Zablah et al. (2004, p.481) confirm that “CRM
is concerned with the development and maintenance of a portfolio of
profit-maximizing customer relationships that is likely to include exchange relationships that vary along the transactional-relational continuum.” This characteristics has two main consequences: it impacts
the target of the CRM initiative and lowers its emphasis on customer
related knowledge:
• Since CRM allows to some extent transactional and non collaborative relationships, its target includes all customers and
prospective customers of the firm. Ryals & Knox (2001, p.535)
confirm that “CRM provides management with the opportunity to implement relationship marketing on a company-wide
basis.” While relationship marketing emphasizes the relationship and interaction with individual customers, CRM provides
the firm with the ability to focus on the entire market. Plinke
(1997, p.19) categorizes CRM as a relationship marketing program targeted on the market or some of its segments. The close
association between IT and CRM is derived from this aspects:
firms rely on technology in order to manage, support, and even
individualize the interactions with customers.
• Since CRM is not focused on the individualization of the value
proposition, it also has a smaller emphasis on customer re-

2.3. Three Approaches Related to Customer Intimacy

55

lated knowledge than customer intimacy. The previously proposed strategy focused definition of CRM does not mention
customer knowledge and its management. In the six previously described CRM subprocesses, customer needs, as a form
of knowledge about the customer, are only mentioned in the
offer and the contract management. These processes, however,
do not detail the management and dissemination of customer
knowledge. Gibbert et al. (2002) argue that CRM is only focused
on knowledge about customers: customer relationship management mines knowledge about customers in order to achieve
customer retention, but does not consider knowledge from customers in order to improve the value proposition for the customer.
The next section summarizes the results of the analysis of the concept
of customer intimacy performed in this chapter.

2.3.4. Customer Intimacy: A Specific Adoption of the
Marketing Concept
In the previous sections, three concepts closely related to, but distinct from, customer intimacy have been introduced: key account
management, market orientation, and customer relationship management. The commonalities and differences between these marketing endeavors and customer intimacy have been outlined and can
be summarized along the following three dimensions, as depicted
in table 2.1: primary objective of the program, focus on customer
relationships, and focus on customer knowledge.
• Primary objective of the program
The first dimension refers to the objective of the marketing initiative. While the primary objective of customer intimacy is to
achieve a competitive advantage through the individualization
of the value proposition and the fulfillment of customer needs,
thereby providing the best solution to the customer, key account management focuses on retaining the most important

56

2. Towards Customer Intimacy

Table 2.1.: Comparison of Customer Intimacy With Other Marketing
Programs
Customer
Intimacy

Key
Account
Management

Market
Orientation

Customer
Relationship
Management

Best
solution (for
all
customers)

Customer
retention
and
customer
value maximization
(for selected
customers)

Profitability
and market
position improvement

Profitability
of the
relationship
portfolio

Focus on
customer
relationships

++

++

−

+

Focus on
customer
knowledge

++

++

++

+

Primary
objective of the
program

customers and on maximizing their value, CRM aims at achieving a portfolio of profitable relationships, and market orientation considers the overall profitability of the firm and its position on the market. Customer intimacy is more focused on
the individualization of the value proposition than key account
management and CRM because the entire customer intimate
organization is structured around the objective to provide a
solution fitting the requirements of the customer, whereas in
the case of key account management and CRM, the individualization of the value proposition is achieved only if this is
necessary to keep the customer and if this is profitable from a
long-term perspective. With regard to market orientation, the
individualization of the offering is perceived as a means to respond to the acquired market intelligence. It is only performed
if it improves the overall profitability of the firm and its position

2.3. Three Approaches Related to Customer Intimacy

57

on the market.
• Focus on customer relationships
The second dimension refers to the establishment of relationships with customers. Since customer intimacy, key account
management, and CRM are all grounded in the concept of relationship marketing, these three concepts focus on the establishment of customer relationships. They are in particular a key
requirement for the successful implementation of customer intimacy and key account management. Market orientation, however, is different and has a lower emphasis on relationships. Relationships are perceived, in the context of market orientation,
as a means to acquire customer knowledge. Indeed, a market
orientation program can be carried out in a transactional perspective.
• Focus on customer knowledge
The third aspect is concerned with the management of customer knowledge. CRM has a lower emphasis on customer
knowledge than key account management and customer intimacy, as it primarily focuses on knowledge about the customer,
and more precisely on mining this knowledge. On the contrary,
key account management and customer intimacy consider customer knowledge as a fundamental source of competitive advantage and develop a stronger emphasis on its management.
Customer knowledge is also a central aspect of market orientation. Market orientation, however, also focuses on knowledge
related to competitors in order to determine the market position of the firm.
In conclusion, customer intimacy can be perceived as a highly developed implementation of the concept of relationship marketing with
a high focus on establishing customer relationships, on managing
customer knowledge, and on leveraging these two aspects in order
to derive competitive advantages. Moreover, its closeness to the main
service dimensions and to the service-dominant logic makes it a very
well suited strategy for all organizations which are going through a
servitization endeavor.

3. Methods and Techniques to
Assess Customer Intimacy
The objective of this chapter is to introduce the methods and techniques leveraged in this thesis in order to perform the assessment of
customer intimacy in a Business to Business (B2B) context, namely
network analysis and data mining.
In order to achieve the objective to provide the customer intimacy
assessment along multiple degrees of granularity, from a focus on the
entire customer organization to a specific analysis of customer teams
and employees, this thesis proposes to apply social network analysis
techniques which provides this ability to consider different entities
and different levels of detail as well as to visualize the information
using graph based representations. Thus, section 3.1 will introduce
the concept of network analysis.
An essential part of this thesis lies in the application of data mining
techniques in order to calibrate and validate the generic customer
intimacy metrics presented in chapter 5. Therefore, section 3.2 will
subsequently outline the main steps of the data mining process as
well as the algorithms chosen in this thesis in order to perform the
analysis.

60

3. Methods and Techniques to Assess Customer Intimacy

3.1. Network Analysis
The application of network analysis and more specifically social network analysis in order to understand relationships among B2B organizations has already been established in past literature. Gummesson
(2008, p.296) argues that network theory is “more comprehensive” in
that regard than other theories such as systems or transaction costs
theories because it does not focus on boundaries between the different actors, but rather on the inter-organizational aspects. Knoke &
Yang (2008, p.1) confirm that the application of social network analysis in the social science literature has grown exponentially over the
past 30 years, and indicate that a significant benefit of social network analysis lies in the consideration of multiple levels of analysis,
defined as “individual and systemic”, which allows an understanding of the “variation in structural relations and their consequences.”
Brandes & Erlebach (2005b) explain that three different levels of analysis are available: element-level analysis, group-level analysis, and
network-level analysis. This characteristic of social network analysis
allows to perform the assessment of customer intimacy at multiple
levels of details, such as individuals, teams and business units, and
whole organizations and, thus, confirms the relevance of social network analysis in this thesis.
Networks and more specifically social networks have been defined
in numerous ways in past literature. An initial contribution to this
definition is provided by Mitchell (1969, p.2) who argues that a social network is “a specific set of linkages among a defined set of
persons, with the additional property that the characteristics of these
linkages as a whole may be used to interpret the social behavior of
the persons involved.” This definition emphasizes the purpose of social network representation, which is to gain a better understanding
of the actors and their relationships. More recently, Knoke & Yang
(2008, p.8) presented a definition which is more focused on the inherent composition of social networks: they define a social network
as “a structure composed of a set of actors, some of whose members
are connected by a set of one or more relations.” In the context of
this thesis, the actors are the provider and customer employees and

3.1. Network Analysis

61

the relations consist of the multiple relationships established through
interactions and shared activities.

3.1.1. Graph Theory for the Representation of Social
Networks
Graph theory has been widely adopted for the representation of social networks as the concepts of actors and relations can easily be
mapped to the graph theory’s notions of vertices and edges. This
thesis adopts the standard graph terminology explained in Brandes
& Erlebach (2005a, p.7): “a graph G = (V, E) is an abstract object
formed by a set V of vertices (nodes) and a set E of edges (links)
that join (connect) pairs of vertices.” Two vertices connected via an
edge are adjacent or neighbors and are called the end vertices of the
edge. It is possible to calculate the degree d(v) of the vertex v by
counting the number of edges in E which have the vertex v as one of
their end vertices. In this thesis, the actors, which are the employees
of the provider and customer organizations are represented by vertices, and the relationships established among them are represented
by edges on the graph. Thus, d(v) is a representation of the number
of direct contacts of the employee v inside the network. Graphs can
be characterized with two additional properties:
• A graph G = (V, E) can be directed or undirected. If the graph
is directed, the order of the end vertices of an edge is relevant
for understanding the graph: the edge eu,v = {u, v} formed by
the end vertices u as origin and v as destination is different
from the edge ev,u = {v, u} whose origin and destination are
respectively the vertices v and u. If the graph is undirected,
the notions of origin and destination to qualify the end vertices
of an edge become irrelevant: the vertices u and v are simply
connected via the edge: eu,v and ev,u have the same meaning in
the graph. In this thesis, the graphs presented in chapter 5 are
undirected as the values of the calculated customer intimacy
components at the individual level do not require to specify
whether the end vertices of the edge are the origin or the destination vertices.

62

3. Methods and Techniques to Assess Customer Intimacy

• A graph G = (V, E) can be weighted or unweighted. If the
graph is weighted, then a numerical value is associated to each
edge on the graph. More formally, the weights can be derived by applying a weighting function ω : E → <. If wi,j is
the weight of the edge ei,j , then wi,j = ω (ei,j ). Since the same
edge may have a high or a low weight depending on the chosen weighting function, this function significantly impacts the
graph representation of the social networks. Moreover, using
the same set of data, an infinite number of weighting functions can be derived (De Choudhury et al., 2010). Therefore, the
weighting function ω has to be carefully determined in order to
achieve the objective of the graph representation. In this thesis,
the graphs which are considered and calculated are weighted.
As detailed in chapter 5, an objective of this thesis is to determine the weighting functions which provide the most accurate
assessment of the values of the customer intimacy components
defined in chapter 4, such as acquired customer knowledge and
established customer relationships.
Two different types of matrices, the incidence matrix and the adjacency matrix provide a formal mathematical representation of the
graph G (V, E) (Brandes & Erlebach, 2005a). In this thesis, the adjacency matrix A( G ) is used by the algorithms which have been
designed for the calculation of the different graphs and their customer intimacy values. The rows and columns of this matrix both
represent the vertices V = {v1 , ..., vn } of the graph, n being the cardinality of V. Thus, A( G ) is a square matrix of size n × n. The
entry a(i, j) in this matrix indicates the existence of an edge in the
graph between the nodes i and j if its value is equal to 1. Otherwise,
its value is equal to 0. The adjacency matrix is defined as follows:
A( G ) = [ ai,j ] | ∀i, j 1 ≤ i, j ≤ n with:

1
if ei,j ∈ E
ai,j =
0
otherwise
Moreover, as described by Newman (2004), since the graphs determined in this thesis are weighted, it is also possible to calculate the

3.1. Network Analysis

63

weighted adjacency matrix W.1 In this matrix, the value of the entry
w(i, j) is equal to the weight of the edge ei,j if the edge ei,j exists, and
to 0 otherwise. With ω : E → < being the function defined to calculate the weights of the edges in the graph G, the weighted adjacency
matrix W ( G ) is defined as follows: W ( G ) = [wi,j ] | ∀i, j 1 ≤
i, j ≤ n with:

ω (ei,j )
if ei,j ∈ E
wi,j =
0
otherwise
This thesis focuses on the representation of the customer intimacy
established between two distinct entities, the provider organization
P and the customer organization C, as well as between their respective employees. Thus, the graph representation of the social network
investigated in this thesis has a specific topology named weighted
bipartite graph. Asratian et al. (1998, p.7) explain that “a graph G
is bipartite if the vertex set V ( G ) can be partitioned into two sets
V1 and V2 in such a way that no two vertices from the same set are
adjacent.” In this thesis, if VP and VC represent the sets of provider
and customer employees, the edges of the graph G (V, E) all have
one end vertex in the set VP and the other one in the set VC : there
is no edge between two nodes which belong to the same set VP or
VC . Figure 3.1(a) illustrates such a bipartite graph representation
with the provider and customer organizations consisting of four and
three employees: VP = { P1 , P2 , P3 , P4 } and VC = {C1 , C2 , C3 }. The adjacency matrix of bipartite graphs has a special characteristic. As explained by Asratian et al. (1998, p.16), “let G be a graph with vertices
v1 , v2 , ..., vn and adjacency matrix A( G ) = [ ai,j ]. Then G is bipartite if
and only if there is a permutation Π of the set {1, 2, ..., n} so that the
matrix A0 ( G ) = [ aΠ(i),Π( j) ] has the following form:


0 B
BT 0
1

the adjacency matrix is sometimes called binary adjacency matrix to
differentiate it from the weighted adjacency matrix (Kiss, 2007, p.72).

64

3. Methods and Techniques to Assess Customer Intimacy

where B T is the transpose of B.” Indeed, as depicted in figure 3.1(b),
the adjacency matrix of the graph proposed in figure 3.1(a) presents
such a structure.
Provider
Provider
PP
P1P1

P2P2

11
22

C1C1

P1P1 P2P2 P3P3 P4P4 C1C1 C2C2 C3C3

P3P3

P4P4

11

0.50.5 0.70.7

C2C2

0.50.5 2 2

C3C3

Customer
CC
Customer

(a) Graph Representation

P1P1 0 0

00

00

00

22

00

00

P2P2 0 0

00

00

00

1 1 0.7
0.7 0.5
0.5

P3P3 0 0

00

00

0 0 0.5
0.5 0 0

00

P4P4 0 0

00

00

00

00

11

22

C1C1 2 2

1 1 0.5
0.5 0 0

00

00

00

0.7 0 0
C2C2 0 0 0.7

11

00

00

00

C3C3 0 0 0.5
0.5 0 0

22

00

00

00

(b) Weighted Adjacency Matrix

Figure 3.1.: A Weighted Bipartite Graph Representation of the
Provider-Customer Relationship

3.1.2. Centrality Metrics for the Analysis of Social Networks
In order to perform an analysis of the social network based on the
graph representation presented in the previous section, various centrality metrics have been proposed in past literature (Freeman, 1979).
Centrality metrics are particularly important for the analysis of networks as they enable an aggregation of the information presented in
the graph and they provide an understanding of the relative position
and importance of each actor inside the network. Many centrality
metrics can be calculated in order to assess diverse characteristics of
a node in the graph (Koschützki et al., 2005). The following three
centrality metrics have been considered in this thesis as they are well
established for understanding the role and importance of each actor
in the social network (Buechel & Buskens, 2008; Freeman, 1979):
1. Degree Centrality
Degree centrality is one of the first centrality metrics which has

3.1. Network Analysis

65

been conceived and is, in its first definition, a synonym of the
previously defined notion of degree (Koschützki et al., 2005).
The degree centrality CD (i ) of the vertex i in the graph G (V, E)
indicates the number of adjacent vertices to i, or the number of
edges which have i as one of their end vertex. Considering the
previously defined adjacency matrix A( G ) of the graph G and
n being the cardinality of V, CD (i ) is calculated as follows:
n

CD ( i ) =

∑ ai,j

(3.1)

j =1

In order to make the degree centrality comparable among graphs
of different sizes, a normalized form of the degree centrality has
been proposed, in which the degree centrality is divided by the
maximum number of potential neighbors on the graph (Freeman, 1979; Wasserman & Faust, 1994). With n being the cardinality of V in the graph G (V, E), the normalized degree cen0
trality CD
(V ) is calculated as follows:
0
CD
(i )

n
∑ j=1 ai,j
=
n−1

(3.2)

The degree centrality and normalized degree centrality are indications of the neighborhood of the actors in the network as
they specify the numbers of actors which can be directly reached.
In this thesis, since the calculated graphs are bipartite, these
two centrality metrics indicate the numbers of relationships
established by a provider (resp. customer) employee inside the
customer (resp. provider) organization.
2. Closeness Centrality
The closeness centrality CC (i ) reflects to which extent the vertex
i is near or far from the other nodes on the graph. Sabidussi
(1966) proposed a first calculation of the closeness centrality
based on the notion of distance di,j between two vertices i and
j. This distance di,j is calculated as sum of weights of the edges

66

3. Methods and Techniques to Assess Customer Intimacy

that belong to the so called geodesic or shortest path that connect i and j. Using this measure, the closeness centrality is
defined as follows:
1
CC (i ) = n
(3.3)
∑ j=1 di,j
As for the degree centrality, a normalized version has been proposed in order to remove the variation due to network size effects (Freeman, 1979; Wasserman & Faust, 1994):
CC0 (i ) =

n−1
n
∑ j=1 di,j

(3.4)

In this thesis, since the customer intimacy graphs are weighted
and bipartite, the geodesic distance between i and j is simply
equal to the weight wi,j of the edge ei,j . Thus, the closeness
and normalized closeness centrality metrics are calculated as
follows:
1
CC (i ) = n
(3.5)
∑ j=1 wi,j
CC0 (i ) =

n−1
n
∑ j=1 wi,j

(3.6)

3. Betweenness Centrality
The third important centrality metric is called betweenness centrality. Its objective is to indicate the relative importance and
power of control of each vertex of the graph. Vertices that have
a high betweenness centrality are located on a high number
of geodesic paths that connect the other nodes in the graph.
Since the graphs considered in this thesis are bipartite, and because the focus is only the relationships between provider and
customer employees, there is no vertex on the graph which is
located on other vertices’ geodesic path. As a consequence, this
metric is not relevant in this thesis and, thus, not further detailed in this section. Additional information on this metric can
be found in Koschützki et al. (2005, p.29).

3.1. Network Analysis

67

3.1.3. Using Social Network Analysis for Assessing
Customer Intimacy
In the previous sections, the notion of a social network, its representation in the form of a graph containing vertices and edges, as
well as its analysis by means of centrality metrics have been explained. In order to perform an analysis of the social network, it
is also necessary to explicit the meaning of the relational ties that
exist between actors in the network, and which are represented by
edges between the vertices of the graph. Wasserman & Faust (1994,
p.18) explain that relationship ties can indicate an extensive number
of meanings such as formal associations, affiliations, behavioral interactions, or evaluations of persons by others. An original aspect of
this thesis lies in the consideration of two different types of relationship ties and in their association by means of data mining techniques
in order to calibrate the model to assess customer intimacy. The two
types of relationship tie considered in this thesis are the following:
• Behavioral interaction
When the relationship ties indicate some behavioral interaction,
the weight of each tie is derived from past communications and
activities that occurred between the two actors related to the
tie. In that case, the data collected to design the social network
consists of past observations or archival records (Wasserman
& Faust, 1994, p.49). Following this approach, it is explained
in chapter 5 how data contained in the provider’s information
system is collected in order to calculate multiple customer intimacy metrics based on behavioral interaction.
• Evaluation of one person by another
When the relationship ties indicate some evaluations of persons
by others, the actors in the social network are asked to answer a
set of questions related to other actors. These questions should
reflect the objective of the social network representation which
is, in this thesis, the assessment of various customer intimacy
components. The data is collected either by means of interviews or through the completion of a questionnaire by the respondents. For scalability reasons, the questionnaire option has

68

3. Methods and Techniques to Assess Customer Intimacy

been chosen and a “customer intimacy questionnaire” has been
conceived in the course of this thesis. While the actual content
of the questionnaire is introduced in chapter 5, the design characteristics of this questionnaire are outlined in the following
paragraphs.
The questions asked to the respondents can either reflect a “complete
ranking” or a “rating” of the relationship ties (Wasserman & Faust,
1994, p.47). In the complete ranking approach, the respondents are
asked to order or to prioritize the different ties on the network with
regard to a specific attribute. For instance, the respondents are asked
to rank the relationships they have established with different customer employees. In the rating approach, the different relationship
ties are considered independently from each other and the respondents are asked to assess the different ties on a certain scale. For
instance, the respondents are asked if their relationships with different customer employees are low, medium, or high. Since “ranking”
the different customers and their employees is out of the scope of
this thesis, the rating approach has been chosen in order to assess
the customer intimacy components.
In order to design this “rating” customer intimacy questionnaire, the
well established approach based on Likert-type scales has been followed. Miller & Salkind (2002, p.330) explain that a Likert-type scale
is a “summated scale consisting of a series of items to which the subject responds.” These items are presented in the form of assertions
for which the respondent evaluates the intensity of his agreement
or disagreement by selecting a value comprised between one and
seven.2 In order to ensure the validity of the Likert-type scales developed in this thesis, the different series of items created to assess the
customer intimacy components have been conceived upon past literature and previously created questionnaires which are mainly rooted
in the field of relationship marketing. These are further detailed in
chapter 5.
2

Some Likert-type scales are based on a different number of intensity grades
such as five, six, or ten.

3.2. Data Mining

69

An important characteristic of Likert-type scales for the rest of this
thesis is the nature of the scale itself. There is indeed some discussion
about whether Likert-type scales should be considered as ordinal or
interval scales.3 As explained by Jamieson (2004), Likert-type scales
are in their essence ordinal, even though several researchers use them
as interval scales. Thus, within the scope of this thesis, the designed
Likert-type scales are considered as ordinal scales. As described in
section 3.2, this characteristic influences the selection of data-mining
algorithms used for calibrating the model.
Further information about social networks, and more specifically
about their actual application in this thesis is provided in chapter 5.
The next section introduces the data-mining approach used in this
thesis for calibrating the assessment of the customer intimacy components.

3.2. Data Mining
Since the calibration of the customer intimacy assessment presented
in chapter 5 and applied in chapter 7 is based on data mining techniques and methods, the objective of this section is to introduce the
underlying data mining concepts which are relevant for this thesis.
Part 3.2.1 introduces the process of “Knowledge Discovery in Databases” (KDD) proposed by Fayyad et al. (1996b), and on which the CI
Analytics methodology elaborated in section 5.1 is aligned. This part
subsequently elaborates on the concepts of data-mining and machine
learning and puts them in relation to the KDD process. Part 3.2.2
motivates the selection of machine learning algorithms considered
in this thesis and shortly describes them. Finally, part 3.2.3 details
the means used for validating data-mining models and, thus, for
confirming the overall approach proposed by this thesis to assess
customer intimacy.
3

An explanation of the difference between ordinal and interval scales is
proposed in Hair et al. (2010, p.5).

70

3. Methods and Techniques to Assess Customer Intimacy

3.2.1. The Process of Knowledge Discovery in Databases
With the exponential increase of data created, stored, and used over
the past decades, in part due to the rise of internet and new information and communication technologies, new solutions have been
required in order to analyze data and to extrapolate some sense out
of it. Thus, the development of solutions, methods, and techniques
for transforming data into actionable and more compact forms of information and knowledge has received considerable interest in both
academia and practice. This overall process of leveraging this data to
generate some knowledge has been called the Knowledge Discovery
in Database (KDD) process. Fayyad et al. (1996a, p.6), who originally
introduced this notion, define it as “the nontrivial process of identifying valid, novel, potentially useful, and ultimately understandable
patterns in data.” Fayyad et al. (1996b, p.37) argue that this process
enables “mapping low-level data into other forms that might be more
compact, more abstract, or more useful.” This process comprises of
six different steps which are depicted in figure 3.2:
1. Problem Definition
The first step in this process consists of obtaining a thorough
understanding of the investigated problem and its context, as
well as in identifying the sources of data which are relevant for
providing a solution. As explained in the introduction, the objective of this thesis is to find some patterns in customer related
data available in the provider’s information system in order to
perform an assessment of the degree of customer intimacy established with different customers.
2. Selection
The second step relates to the selection of the data records on
which the analysis will be completed and to the identification
of the actual fields in the data set that will be considered.
3. Pre-Processing
The third step concerns cleaning the data, such as removing
noise and outliers which may prevent from identifying the patterns, and handling the missing values in the data set.

3.2. Data Mining

71

4. Transformation
In the fourth step, the data is transformed in order to emphasize its most important characteristics. This involves the aggregation of the variables in the data set to create summated scales
as well as the projection of the data in orthogonal dimensions
in order to reduce the number of variables.
5. Data Mining
The fifth step refers to the analysis of the data itself through the
application of various machine learning algorithms. This step
is further detailed in the next paragraph.
6. Interpretation/Evaluation
Finally, the last step consists of the validation of the model in
order to ensure that it can be used with other data sets as well as
in its interpretation in order to derive some theoretical or practical knowledge. This step is further detailed in section 3.2.3.
Interpretation/
Evaluation
Data-Mining
Preprocessing

Transformation
Knowledge

Selection
Problem
Definition

Patterns

Data

Target
Data
Set

Transformed
Data
Preprocessed
Data

Figure 3.2.: The Knowledge Discovery Process (Fayyad et al., 1996b)
Within the KDD process, the fifth activity is concerned with the analysis of the data itself and more precisely with the detection of patterns among the multiple data records. This aspect is referred to
as data-mining (Witten et al., 2011). Fayyad et al. (1996b, p.41) confirm that “data mining is a step in the KDD process that consists of
applying data analysis and discovery algorithms that produce a particular enumeration of patterns (or models) over the data.” Notably,

72

3. Methods and Techniques to Assess Customer Intimacy

since data-mining is certainly the most significant activity of the KDD
process, the concepts “knowledge discovery in database” and “data
mining” are sometimes used as synonyms (Kiss, 2007, p.12, Mitchell,
1999). For instance, the Cross Industry Standard Process for Data
Mining (CRISP-DM) consists of 6 steps4 which are to a high extent
aligned to the KDD process (Chapman et al., 2000).
Several computer-science based methods and algorithms have been
conceived in order to perform the analysis of the data. These are
called machine learning algorithms. As explained by Alpaydin (2010,
p.3), “machine learning is programming computers to optimize a
performance criterion using example data or past experience [...]
Their application to large databases is called data mining.” Such algorithms are rooted in the field of artificial intelligence as they have
to be able to adapt to changing environments. The principle of machine learning is that the algorithm is applied to a set of data records
called training set in order to create a model consisting of multiple
patterns which present a structural description of the data set (Witten et al., 2011, p.8). After being validated, this model can be applied
on other data sets in order gain new insight. There are two different
types of machine learning algorithms:
• Unsupervised Learning
In the case of unsupervised learning, no specific field in the
data is considered as a reference: all fields are input data and
the objective is simply to identify regularities in the input (Alpaydin, 2010, p.11).
• Supervised Learning
In the case of supervised learning, an attribute of the dataset
is considered as the target or as the output variable of the algorithm. The algorithm is applied on the training set in order
to “learn” the value of this attribute based on all other fields,
which are called the input variables of the algorithm: “the task
is to learn the mapping from the input to the output” (Alpaydin, 2010, p.9). Classic types of supervised learning are regres4

These steps are (1) Business Understanding, (2) Data Understanding, (3)
Data Preparation, (4) Modeling, (5) Evaluation, (6) Deployment.

3.2. Data Mining

73

sion and classification. If the target variable on the dataset is
numeric and continuous, then a regression is performed: the
supervised algorithm aims at creating a model which predict
as closely as possible the value of the target field, based on the
available input fields. If the target variable is nominal or ordinal, then a classification is performed: the algorithm aims at
creating a model that predict the class or the order of the record
specified in the target field, based on the other input fields.
In this thesis, the objective is to use the customer related data available in the provider’s information system in order to predict the customer intimacy component values which have been empirically assessed. Consequently, the supervised machine learning approach is
followed as the target variable is derived from the empirical results.
As explained in section 3.1.3, this empirical analysis of the customer
intimacy components is performed using ordinal Likert-type scales.
Thus, from a machine learning perspective, the aim of this thesis is
to perform a classification.

3.2.2. Selection of the Machine Learning Algorithms
Four classification algorithms have been considered in this thesis.
While the first part of this section motivates their selection, the remaining parts briefly describe them.
3.2.2.1. Choosing Relevant Algorithms
Multiple machine learning algorithms are available in order to solve
a classification problem. Many of them use, to different degrees,
concepts rooted in classic inferential statistics and in Bayesian decision theory.5 Indeed, Witten et al. (2011, p.28) confirm that there
is no strict difference between machine-learning and statistics but
“a continuum of data analysis techniques.” However, in contrast to
classic inferential statistics which require the dataset to fulfill certain
5

The Bayesian decision theory focuses on the estimation of class probabilities,
knowing certain conditions apply or certain observations were
made (Alpaydin, 2010, p.3, p.48).

74

3. Methods and Techniques to Assess Customer Intimacy

conditions, such as normality, homoscedasticity, and linearity, many
machine learning algorithms have been designed for data contained
in databases, which in most cases violate these conditions (Hair et al.,
2010, p.72, Press, 2003, p.6). The algorithms which can be applied to
data whose distribution is unknown are called non-parametric algorithms (Alpaydin, 2010, p.164). Since it cannot be assumed that the
customer related data available in the provider’s information system
follows a specific distribution, only non-parametric machine learning
algorithms are considered in this thesis.
Several factors influence the performance of a machine learning algorithm on a specific data set such as the number of target classes,
the distribution of the target class, the total number of cases and attributes, and the average number per class (Nisbet et al., 2009, p.257).
Moreover, there is no absolute analytical rule for determining the
most relevant algorithms upon certain characteristics of the dataset
(Kalousis et al., 2004, Kiss, 2007, p.23). Thus, several projects were
conducted over the past decades in order to empirically assess the
performance of data-mining algorithms. In this thesis, the machine
learning algorithm selection was performed on the basis of the results from three different analyses:
• The STATLOG project is considered as one of the most exhaustive evaluation of data mining algorithms as it compares
the performance of 20 classification methods on 20 different
datasets (Michie et al., 1994). Some of its conclusions are as follows: (i) the nearest neighbor algorithm performed very well on
all datasets, even though it was the slowest on large datasets;
(ii) the neural network with back-propagation algorithm obtained the highest or near highest predictive performance in
nearly all cases; (iii) all decision trees had a fairly constant “average” performance across all datasets.
• Lam et al. (2002) benchmarked on 50 data sets their custommade algorithm “ICPL” with the algorithms k-nearest neighbor,
C4.5 decision tree and support vector machine. The k-nearest
neighbor algorithm achieved the highest classification accuracy

3.2. Data Mining

75

followed by the support vector machine and the ICPL algorithms.
• Ali & Smith (2006) analyzed the performance of eight algorithms (classifiers) on 100 different datasets. No algorithm could
be identified whose performance was constantly above average
for all 100 datasets. The support vector machine algorithm obtained the best accuracy. The decision tree C4.5 and the neural
network algorithms also obtained very good results in terms on
percentage of correctly classified instances.6
Based on this analysis, it appears that the following algorithms are
highly relevant classifiers:
1. Decision tree C4.5
2. k-nearest neighbor
3. Neural network with back-propagation
4. Support vector machine
Thus, these four algorithms have been considered in the scope of this
thesis. The next parts of this section present further details on each
of them.
3.2.2.2. Decision Tree C4.5
The machine learning algorithm C4.5 belongs to the family of decision tree classifiers, which have the advantage of being graphically
representable and, thus, easily interpretable. A decision tree is “a hierarchical data structure implementing the divide-and-conquer strategy” (Alpaydin, 2010, p.187). Considering a certain data record in
the database with multiple attributes, the decision tree models the
classification task in multiple sequences of tests on the attributes in
order to determine or predict the class of the record. The different
6

The performance indicators accuracy and percentage of correctly classified
instances are developed in section 3.2.3.2.

76

3. Methods and Techniques to Assess Customer Intimacy

test sequences which lead to the classification are hierarchically represented in the form of a tree,7 which consists of one root node, internal nodes, branches, and terminal leaves. In a decision tree, the nodes
– also called test nodes – represent the attributes on which the tests
are applied, the branches represent the different test predicates, and
the terminal leaves constitute the possible classes. Dunham (2002,
p.93) proposes the following formalization of a decision tree: “Given
a database D = {t1 , ..., tn } where ti = {ti1 , tih } and the database
schema contains the following attributes A = { A1 , A2 , ..., Ah }. Also
given a set of classes C = {C1 , C2 , ..., Cm }. A decision tree or classification tree is a tree associated with D that has the following properties: (i ) each internal node is labeled with an attribute Ai ; (ii ) each
arc is labeled with a predicate that can be applied to the attribute
associated with the parent; (iii ) each leaf node is labeled with a class
Cj .”
The objective of decision tree classifiers, such as the C4.5 algorithm,
is to induce the decision tree, which means to determine the best
way of splitting the data and, thus, to identify effective and accurate
sequences of tests on the attributes in order to assess the class of
the different records (Tan et al., 2006, p.151). C4.5 was proposed by
Quinlan (1986) as a successor of the ID3 algorithm. It uses the information gain ∆in f o in order to infer the decision tree. This information
gain represents the potential increase in the information value of the
decision tree that would result from extending it with an additional
sub-tree. This sub-tree indicates that an additional test is required
in order to lead to the classification decision. More formally, the information value is called entropy Iin f o and it measures the degree of
purity of the different nodes in the tree (Tan et al., 2006, p.158).8 If m
represents the number of classes, t a node in the tree, and p(i |t) the
fraction of records belonging to a class i at the given node t, then the

7
8

A tree is a special type of graph that fulfills the following two conditions: it
is connected and it is acyclic (Wasserman & Faust, 1994, p.119).
Other impurity measures include Gini and classification error.

3.2. Data Mining

77

entropy Iin f o (t) is defined as follows:
m

Iin f o (t) = − ∑ p(i |t)log2 ( p(i |t))

(3.7)

i =1

If Tchildren = {t1 , t2 , ..., tk } represents the set of children nodes of the
node t, and N (ti ) the number of records associated to the node ti ,
then the information gain ∆in f o is calculated as follows:
k

N (t j )
× Iin f o (t j )
N
(
t
)
j =1

∆in f o = Iin f o (t) − ∑

(3.8)

In order to infer the decision tree, the algorithm C4.5 creates the
different nodes of the tree in an iterative manner, starting with the
root node. To create the node ti , the algorithm evaluates the potential information gain ∆in f o obtained with each input attributes. The
attribute with the highest gain is set to the test node ti . This operation is then reapplied in order to determine the children nodes of ti
and so on, until a stop criterion such as the maximum tree depth or
the minimum number of items per class is reached (Tan et al., 2006,
p.164).
3.2.2.3. k-Nearest Neighbor
The k-nearest neighbor algorithm belongs to the so called “lazy learners” or “instance-based learning classifiers” as it does not create an
explicit model representation of the knowledge provided in the training data set (Tan et al., 2006, p.223, p.226). Instead, the different
records contained in the training data set are all memorized by the
algorithm. When a new record r has to be classified, the algorithm
calculates its distance to all records in the training set, the shortest
distance indicating the highest degree of similarity. The class of r
is then determined upon the classes of its k-nearest neighbors, for
instance using a majority vote scheme.
More formally, considering a training data set D = {d1 , d2 , ..., dn } of
size n whose instances have the set of attributes A = { x1 , x2 , ..., xh }

78

3. Methods and Techniques to Assess Customer Intimacy

of size h, each record di can be represented by a point in the hdimensional space Rh . In order to assess the class of the new record r,
r is also represented as a point in the space Rh and its Euclidean distance to all items in D is calculated. Then, the list Dr of the k-nearest
neighbors of r is computed and their respective classes is reviewed. If
the k-nearest neighbors belong to the same class C1 , then r is also set
to C1 . If the k-nearest neighbors belong to different classes, then majority vote or distance-weighted voting schemes are applied in order
to assess the class of r.9
Even though this algorithm has proven its effectiveness, a key challenge in the k-nearest neighbor algorithm resides in the appropriate
determination of the number k. If k is chosen too small, there is a
risk of misclassifying a record because of its closeness to one specific
noisy item in the training set.10 If k is chosen too large, then some
items in the training set which are far from r and of different class
may remain influential in the classification of r if they belong to the
k-nearest neighbors of r.
3.2.2.4. Support Vector Machine
The support vector machine classifier belongs to the kernel machine
learning algorithms (Alpaydin, 2010, p.309). It can be seen as an evolution of statistical learning theory which includes concepts derived
from instance based learning (Witten et al., 2011, p.192). Its strength
lies in its ability to handle high-dimensional data and to consider
both linearly and non-linearly separable data (Tan et al., 2006, p.256).
This algorithm discriminates training records pertaining to two different classes by using a subset of the training data set which are called
the support vectors.
Considering a training data set D = {d1 , d2 , ..., dn } of instances which
have the set of attributes A = { x1 , x2 , ..., xh } and a set of two classes
C = {C1 , C2 }, these instances can be represented as points in the hdimensional space Rh . In order to identify the support vectors, the
9
10

The majority vote and the distance-weighted voting are explained in Tan
et al. (2006, p.226).
Such a problem is called over-fitting.

3.2. Data Mining

79

support vector machine algorithm determines the maximum margin
hyperplane.11 If the data is linearly separable, an infinite number
of hyperplanes can be identified in Rh which discriminate the items
in D of class C1 from those of class C2 by varying the coefficients
of the equation that determines the hyperplane. The support vector
machine aims at identifying the hyperplane whose distances to the
nearest items of class C1 and of class C2 are maximized. As the sum
of these two distances is called the margin of the hyperplane, the
objective of the support vector machine algorithm is to determine
the maximum margin hyperplane. Indeed, it has been established
that “decision boundaries with large margins tend to have better generalization errors than those with small margins.” (Tan et al., 2006,
p.257). Thus, the maximum margin hyperplane should be a better
discriminant of both classes than any other hyperplane.
In the case that the data is not linearly separable, and therefore no
hyperplane can be found in Rh to discriminate the items of class C1
from those of class C2 , it is possible to perform a non-linear transformation of the space Rh and then identify the maximum margin
hyperplane in this newly created space. This operation is, however,
resource intensive and the transformation function is unknown (Tan
et al., 2006, p.272). The use of kernel functions, which “replace the
transformation functions” provides the ability to search for the maximum margin hyperplane in a non-linear model directly into the original space Rn . Further details on the kernel functions are provided
in Alpaydin (2010, p.320).
3.2.2.5. Artificial Neural Network – Multilayer Perceptron with
Back-Propagation
Artificial neural networks are parallel information processing systems which aim at reproducing the mechanisms of biological neural
11

Ostaszewski (1990, p.123) defines an hyperplane as an affine set of
dimension n − 1 in the n-dimensional space <n which divides <n into two
half-spaces. Given a set of items S in <n and a a boundary point (support
vector) in S, the hyperplane H supports the set S in <n at the point (support
vector) a if: (i ) the point a belongs to H and (ii ) S is entirely contained in
one of the two half-spaces formed by H(Ostaszewski, 1990, p.129).

80

3. Methods and Techniques to Assess Customer Intimacy

networks. Similarly to real neural systems in which neurons are connected to each others via axons and synapses, artificial neural networks are composed of multiples neurons or nodes which are interconnected with weighted and directed links (Tan et al., 2006, p.248).
In this thesis, the multilayer perceptron neural network algorithm is
considered. The simpler single layer perceptron consists of a layer
of input neurons which represent the attributes assessed in the classification task, one output neuron whose role is to predict the class,
and directed weighted edges that connect the input neurons to the
output neuron. In order to classify a record r characterized by the
set of attributes A = { x1 , x2 , ..., xh }, the input neurons of the perceptron transfer concurrently r’s attribute values to the output node via
the corresponding weighted edges. The output node computes the
value of the perceptron y as the weighted sum of the inputs. Then,
it uses an activation function s to transform y into a boolean value
which can be associated to a specific class. The activation function s
can be linear, sigmoid (logistic), or based on a threshold value. For
instance, if y ≥ 0 then s(y) = 1 and the record r is classified in C1 . If
y < 0, s(y) = 0 and r is classified in C2 . The learning algorithm of the
perceptron consists in feeding the network iteratively with the items
of the training data set and in adjusting the weights of its edges until the classes predicted by the output node correspond to the actual
classes of the training items.
Since the perceptron only has one single layer and its output neuron
estimates the class based on a weighed sum of the input attributes,
it uses only linear discriminants in order to perform the classification task. In order to remedy this limitation, the multilayer perceptron contains additional intermediate or “hidden” layers of neurons
between the input neurons and the output neurons which provide
the ability to use non-linear discriminants (Alpaydin, 2010, p.246).
Figure 3.3 illustrates such a network in which the classification is
performed upon three input attributes x1 , x2 and x3 . In this example the multilayer perceptron contains one hidden layer composed of
two neurons. In order to classify the record r, its attribute values are
presented to the respective input neurons I1 , I2 and I3 . The hidden
neurons H1 and H2 compute the weighted sums of values delivered

3.2. Data Mining

81

by the input neurons using the weights wi,j indicated on the graph
and transform them with their respective activation function s0 and
s1 . The outputs of the hidden neurons are then passed through to the
output neuron O. This neuron repeats the operation of calculating
the weighted sum with the appropriate weights and of transforming
the value with its own activation function s0 . This value is finally
used in order to assess the class of the record r.
x1

I1
w11
w12

x2

I2

w1

w21

w22
w31

x3

H1
Σ s1
H2
Σ s2

O
Σ s0

y

w2

w32
I3
Input
Layer

Hidden
Layer

Output
Layer

Figure 3.3.: Illustrative Multilayer Perceptron
The multilayer perceptron is a feed-forward neural network with
back-propagation of the error estimate. The feed-forward characteristic indicates that the neural network is unidirectional. Indeed, as
illustrated in figure 3.3, the neurons are only connected to neurons
in subsequent layers which are closer to the output node (Tan et al.,
2006, p.251). The back-propagation of the error estimate feature indicates that the training algorithm of the multilayer perceptron is composed of multiple iterations of the following two phases: during the
forward phase, the training sample records whose class are known
are passed through the network iteratively in order to estimate the
weights of the edges. During the backward phase, the error estimated on the sample records are transferred back to the neurons in
the previous layers in order to adjust the weights. These two phases
are repeated until an acceptable error estimate is reached (Tan et al.,
2006, p.254).

82

3. Methods and Techniques to Assess Customer Intimacy

3.2.3. Evaluation of the Machine Learning Models
Once a machine learning algorithm has been trained to resolve the
classification task, a machine learning model is created. This model
has to be evaluated in order to ensure its ability to classify records
that do not belong to the training data set and, thus, to determine its
generalization error (Tan et al., 2006, p.186). Different methods have
been conceived in order to perform this evaluation such as holdout technique, the bootstrap, and the cross-validation. The first part
of this section introduces these different options and motivates the
choice to use the cross-validation technique in this thesis. In addition,
several criterion have been defined in order to quantify the evaluation, like the precision and recall values or the kappa statistic. The
second part of this section develops the indicators which are used to
evaluate the machine learning models presented in chapter 7.
3.2.3.1. Different Options to Split the Data Set
In order to assess the capability of a machine learning algorithm to
perform a certain classification task, four main techniques have been
proposed (Tan et al., 2006, p.186):
• Holdout Method
The holdout method simply consists of splitting the data set in
two subsets: a training set and a test set. Once the learning
process has been performed on the training set, the resulting
model is applied on the test set. The results achieved by the
model on the test set are used to assess the capability of the
model to determine the actual classes of records that do not
belong to the training set (Tan et al., 2006, p.149).12
• Random Subsampling
The random subsampling method consists of repeating the holdout method several times: If k subsamples are created, the data
set is randomly split k times, resulting in k couples of training
12

In some cases the original data set is split in a training set to create the
model, a validation set to optimize it, and a test set to assess its
performance (Witten et al., 2011, p.149).

3.2. Data Mining

83

and test sets. The machine learning algorithm is, thus, trained
and assessed k times. The overall performance of the algorithm
is calculated as the average performance of the k generated
models on the test sets.
• Bootstrap
Similarly to random subsampling, the bootstrap method generates multiple samples from the original data set, and train and
test the machine learning on each of these samples. Its specificity is that it uses a subsampling with replacement technique
in order to create the sampled training data sets: any record in
the original data set can be selected multiple times to compose
the training set of each sample. The corresponding test set is
then formed by the remaining records of the original data set
which do not belong to the training set.
• k-Fold Cross-Validation
The cross-validation technique is an evolution of the random
sampling method which ensures that all records of the original
data set are allocated the same number of times to the training
sets and exactly once to the test sets: if k samples consisting of a
training set and a test set are generated out of the original data
set D, all records in D are allocated k − 1 times to the training sets and once to the test sets. This constraint ensures that
all potential patterns in the original data set are represented in
both the training and test sets. On the contrary to the bootstrap method, cross-validation does not use subsampling with
replacement.
The cross-validation method has been chosen in this thesis as it is
recognized as “the standard way for measuring the error rate of a
learning scheme on a particular data set” (Witten et al., 2011, p.154).
Indeed, the other techniques all present some drawbacks: the holdout method requires a large amount of data in order to ensure that
both the training set and the test set contain sufficient representative
samples. Moreover, the repartition of the data in both sets has to be

84

3. Methods and Techniques to Assess Customer Intimacy

performed thoroughly since it may influence the evaluation results.13
With the random subsampling method, the bias induced from an incorrect repartition of the data in the training and test sets is removed,
but there is no control on how often records are allocated to the training and test sets, leading to a misinterpretation of the identified patterns. Finally, the bootstrapping method is particularly efficient on
data set of small size. However, Witten et al. (2011, p.156) argues
that the estimation of the error using this method is, in many cases,
overly optimistic. Kohavi (1995) compared the bootstrap and the
cross-validation methods on six different data sets and concluded
with the recommendation to use the “10-fold cross-validation”, in
which the parameter k is set to the value 10.
In order to implement the k-fold cross-validation, the original data
set D is partitioned in k mutually exclusive subsets of equal size and,
thus, k samples are generated. The ensemble of generated subsets is
defined as R = { R1 , R2 , ..., Rk } and the ensemble of generated samples is denoted as S = {S1 , S2 , ..., Sk }. Each sample consists of test
set and a training set. The test set testi and the training set traini of
the sample Si are composed respectively of the records of the part Ri
and of all records which are not in Ri :
Si = {testi , traini } with testi = Ri and traini = D − Ri

(3.9)

In order to evaluate the performance of the machine learning algorithm, the algorithm is trained and assessed on each of the k samples.
The results achieved by the trained models on the k test sets are then
combined in order to determine the overall accuracy of the algorithm.
In this thesis, the parameter k has been set to 10, as recommended
in past literature (Alpaydin, 2010, p.487, Witten et al., 2011, p.153,
Kohavi, 1995). Thus, the original data set has been partitioned in
10 different parts and 10 samples have been generated. Moreover,
as recommended by Witten et al. (2011, p.154), in order to ensure
13

The machine learning algorithm may have a poor or a high performance
depending on whether the patterns identified in the training set also exist in
the test set or not.

3.2. Data Mining

85

that the data partitioning does not bias the evaluation of the performance of the different machine learning algorithms, the entire kfold cross-validation process has been repeated 10 times, each time
with a different partitioning of the original data set: a “10 times 10fold cross-validation” has been performed in order to evaluate the
performance of each configuration of the different machine learning
algorithms.
3.2.3.2. Model Evaluation Criteria
Several indicators have been conceived in order to assess the ability
of a machine learning algorithm to solve a classification problem on
a certain data set. Many of these indicators are derived from the
confusion matrix. Considering a two-class classification task with the
potential classes C1 and C2 , this matrix is a 2 × 2 matrix as depicted in
figure 3.4. The results achieved by the trained model on the records
in the test set are sorted in four categories. If the class of the record
is predicted as C1 and is actually C1 , this record is classified as a
true positive (tp). If the class of the record is predicted as C2 and is
actually C2 , this record is classified as true negative (tn). If the class
of the record is predicted as C1 , but the record belongs in fact to the
class C2 , the record is classified as false positive ( f p). Finally, if the
class of the record is predicted as C2 , but the record in fact belongs
to the class C1 , the record is classified as a false negative ( f n). The
confusion matrix then reports the number of records in the test set
that belong to the different categories.
Predicted Class

Actual
Class

Class C1

Class C2

Class C1

true positive (tp)

false negative (fn)

Class C2

false positive (fp)

true negative (tn)

Figure 3.4.: Confusion Matrix
Using this confusion matrix, multiple indicators have been proposed
in order to assess the performance of a machine learning algorithm.

86

3. Methods and Techniques to Assess Customer Intimacy

The following five indicators are considered as references and are
used in the context of this thesis (Alpaydin, 2010; Witten et al., 2011):
• Success Rate
The success rate indicates the proportion of correctly classified
records in the test set considering both classes C1 and C2 .14 The
success rate is calculated as follows:
Success Rate =

tp + tn
tp + tn + f p + f n

(3.10)

• Precision
The precision, also called accuracy, indicates to which extent
the classification performed by the machine learning model corresponds to reality. It is calculated as the proportion of records
of class C1 among all records which have been classified as C1
by the machine learning model:15
Precision =

tp
tp + f p

(3.11)

• Recall
The recall measure indicates to which extent the machine learning model is capable of retrieving the items that actually belong to the class C1 . It is calculated as the proportion of items
that have been classified as C1 by the machine learning model
among all items which actually are of class C1 :16
Recall =

tp
tp + f n

(3.12)

• F-Measure
14
15
16

Respectively, the error rate can be calculated as the proportion of incorrectly
classified records
The precision can also be calculated for the class C2 . Its calculation is then:
tn
tn+ f n .
tn
The recall can also be calculated for the class C2 . Its calculation is then: tn+
fp

3.2. Data Mining

87

The F-Measure is a combination of the precision and recall values and it is calculated as their harmonic mean (Witten et al.,
2011, p.175):
Recall × Precision
F = 2×
(3.13)
Recall + Precision
• Kappa Statistic
The Kappa Statistic compares the success rate obtained by a
specific machine learning algorithm with the success rate achieved by a “random” algorithm that would randomly allocate the
records in the test set to the class C1 and C2 with respect to the
actual proportion of items of class C1 and C2 in the test set. The
Kappa Statistic is then calculated as the performance increase
between both success rates (Witten et al., 2011, p.166).
100

True Positive (%)

80

60

40

20

0
0

20

40

60

80

100

False Positive (%)

Figure 3.5.: ROC Curve
Finally, in order to graphically represent the performance of the different machine learning algorithms applied in chapter 7, the “ROC

88

3. Methods and Techniques to Assess Customer Intimacy

curve” is used in this thesis. ROC means “Receiver Operating Characteristic.” This graphical technique provides a representation of the
true positive rate as a function of a the false positive rate, both presented as a percentage (Witten et al., 2011, p.174). The ROC curve
provides the ability to visualize the trade-off between these two parameters performed by different machine learning algorithms. For
instance, the ROC curve illustrated in figure 3.5 shows that in order
to achieve a true positive rate of 40%, the false positive rate will be
equal to 10%. However, in order to achieve a true positive rate of
60%, the false positive rate will be much higher and equal to 65%.
This means that this algorithm is efficient if the objective is to select
samples with 40% of true positive records, but inefficient if the objective is to select samples with 60% of true positive records. The x
and y axis on the ROC curve are calculated as follows:
fp
f p + tn
tp
y = True Positive Rate = 100 ×
tp + f n

x = False Positive Rate = 100 ×

(3.14)
(3.15)

Part II.

Conceptual Model

4. Customer Intimacy
Breakdown Analysis
In chapter 2, the value discipline customer intimacy has been explained and put in relationship to other marketing concepts such as
relationship marketing, key account management, and the servicedominant logic. The objective of this chapter is to establish how this
concept can be broken down in multiple component parts, laying
thereby the foundation of the overall model to assessing and monitoring customer intimacy. Based on an analysis of the definition of
customer intimacy and of the constraints of the B2B context, various customer intimacy components have been determined at both
organizational and individual levels. This chapter will develop this
analysis, specify in detail each of these components, and motivate
their relevance for the assessment of customer intimacy.
Section 4.1 will analyze existing approaches for assessing customer
intimacy and will outline the distinctive characteristics of the proposed approach. Section 4.2 will subsequently elaborate on the performed customer intimacy breakdown analysis upon which the customer intimacy model proposed by this thesis is derived. This model
consists of two parts, namely the acquired customer intimacy and the
leveraged customer intimacy. Section 4.3 will detail the components

92

4. Customer Intimacy Breakdown Analysis

pertaining to the acquired customer intimacy and section 4.4 will
develop the leveraged customer intimacy components.

4.1. Existing Approaches for Assessing
Customer Intimacy
The measurement of customer intimacy has been a research topic
addressed from multiple perspectives over the past years. In order to
classify the different solutions proposed in existing literature, three
criteria have been considered:
• Analysis Level
Several degrees of analysis should be considered in order to
thoroughly assess the degree of customer intimacy. While a
general analysis of the activities involving both the customer
and the provider at the organizational level is required, such
as projects and sales contracts, a more detailed perspective focusing on the interactions occurring between the provider and
customer employees is also needed to precisely estimate which
employees and which teams in the provider organization have
become “customer intimate”. Thus, the customer intimacy assessment should be performed at the organizational level as
well as at the individual level.
• Assessment Focus
In this thesis, the objective is to assess the degree of customer
intimacy established with different customers. The focus of the
assessment is, therefore, on customers and more specifically on
the interactions, activities, and projects involving the different
customers. There are, however, other approaches to assess customer intimacy that take a different perspective and focus on
the internal ability of a firm to implement a customer intimacy
strategy. The assessment focus can therefore be on customers
or on the provider organization itself.
• Assessment Type
Two different approaches for measuring customer intimacy have

4.1. Existing Approaches for Assessing Customer Intimacy

93

been investigated in past literature: the analytical approach
which focuses on creating some key indicators out of existing
data and the empirical approach which uses employees’ feedbacks by means of questionnaires and interviews.
Different solutions to assess customer intimacy have been reviewed
in the scope of this thesis. A selection of the most relevant ones
as well as their categorization along the three criteria analysis level,
assessment focus, and assessment type is provided in table 4.1. These
solutions are detailed in the next paragraphs.
Cuganesan (2008) examines the use of financial data to calculate customer intimacy at the organizational level. Based on a case study
with a wholesale financial service company, he suggests two modes
of calculation which differs in the way customer intimacy is enacted: a “sales calculation network” approach and a “numeric calculation network” approach. The sales calculation network approach
is driven by relationships, sales, and business units managers and
focuses on the generation of knowledge about the interests of customers. The numeric calculation network approach is driven by the
market intelligence department and focuses on the creation of performance measures based on market research. However, no details
are provided on how these approaches are actually calculated.
In a balanced scorecard evaluation, Niven (2002) proposes five attributes which can be developed in order to measure customer intimacy. These are customer knowledge, offered solutions, penetration, culture of driving client success, and relationships in the long
term. The operationalization and detailed implementation of these
attributes, however, remain open.
Kaplan (2005, p.1) suggests that “for a differentiated customer intimacy strategy to succeed, the value created by the differentiation
– measured by higher margins and higher sales volumes – has to
exceed the cost of creating and delivering customized features and
services.” Later, he suggests to utilize the time driven activity based
costing introduced in Kaplan & Anderson (2007) in order to assess
these costs and evaluate customer profitability.

94

4. Customer Intimacy Breakdown Analysis

Table 4.1.: Overview of Existing Approaches Towards the Assessment of Customer Intimacy
Analysis Level
Reference Organization Individual

Assessment Focus
Customer

Intern

Assessment Type
Empirical

Analytical

Cuganesan
(2008)

X

X

X

Niven
(2002)

X

X

X

Kaplan
(2005)

X

X

X

Industry
Directions &
IBM
(2006)

X

X

Potgieter
& Roodt
(2004)

X

Tuominen
et al.
(2004)

X

Abraham
(2006)

X

X

X

Yim et al.
(2008)

X

X

X

X

This
thesis

X

X

X

X

X

X

X

X
X

X

4.1. Existing Approaches for Assessing Customer Intimacy

95

An executive report suggests that services provide the opportunity
for industrial companies to significantly deepen the level of customer
intimacy and increase customer control, but it does not explain how
to evaluate this level of customer intimacy and, thus, how to measure
the improvement through the added services (Industry Directions &
IBM, 2006).
Potgieter & Roodt (2004) provide a model in which they consider
customer intimacy from the internal perspective and they conceive
a questionnaire for the assessment of the customer intimacy culture
of an organization. This questionnaire was validated by an empirical
study in a company from the entertainment industry. Their approach
does not consider the actual intimacy achieved with individual customers, but the ability of an organization, and more specifically its
cultural aspects, to support a customer intimacy strategy.
Tuominen et al. (2004) provide a six-layer approach for evaluating
customer intimacy: they differentiate whether the organization (1)
was involved in the customer’s planning process, (2) involved customers in their planning process, (3) partnered and jointly planned
with customers, (4) aligned each other’s operating processes, (5) designed operational interfaces, and (6) formalized the system of joint
decision making. They use this scale to correlate the degree of customer intimacy with the degree of market orientation of the firm
and its internal market intelligence capability, and recognize the importance of partnership and collaboration in the development of a
customer intimacy strategy. However, only a few details are provided on actual implementation, and this solution solely focuses on
the organizational level.
Abraham (2006, p.1) emphasizes the importance of the relationships
between employees. He explains that customer intimacy represents
“the formal or informal set of relationships established between supplier and customer, with a diverse array of partners, from corporate
leadership to functional leadership (engineering, marketing, operations, maintenance, or service) and end-users of products or services.” These dynamic relationships provide multiple points and
frequency of contacts between the company and its customer, as well

96

4. Customer Intimacy Breakdown Analysis

as multiple points of view about the relationship and its benefits to
both parties. According to his work, increasing customer intimacy
can be achieved by improving the attitude of the employees dealing
with the customer.
Yim et al. (2008) propose a model in which they consider both the
customer-staff and customer-firm interactions in parallel. They define intimacy as the bondedness and connectedness of a relationship
between two individuals and investigate how intimacy and passion
can enrich customer service interactions and impact the customerfirm relationship. They validate this model by means of two empirical studies and conclude in particular that customer-staff affection
influences customer-firm affection and customer-firm affection has a
mediating role in strengthening customer loyalty.
This literature review outlines the distinctiveness of the approach
proposed by this thesis. Indeed, as depicted in table 4.1, most of the
existing solutions focus on the organizational level of analysis and
do not consider the degree of customer intimacy established among
employees. This thesis, on the contrary, considers both the organizational and the individual levels of analysis. Then, similarly to several other solutions, this thesis focuses on the actual assessment of
the degree of customer intimacy established with different customers
rather than on the inherent ability of an organization to pursue the
customer intimacy value discipline. This thesis in addition combines
an analytical customer intimacy measurement with an empirical assessment in order to validate the proposed solution.

4.2. Overview of the Customer Intimacy
Breakdown Analysis
This section sets out the overall model to break down customer intimacy into multiple components. Many different aspects should be
considered when developing a model to assess the degree of customer intimacy between a company and its customers. Liljander
& Strandvik (1995) identified within their service relationship quality model that some of these aspects are at the organizational level,

4.2. Overview of the Customer Intimacy Breakdown Analysis

97

while others are at the individual or employee level. Based on this
premise, and in order to achieve the benefits outlined chapter 1, the
model proposed by this thesis intends to include an assessment of the
degree of the customer intimacy established with customers at both
the organizational and individual levels. On the one hand, the individual level of analysis refers to an assessment focusing on customer
and provider employees considered on an individual basis. On the
other hand, the organizational level of analysis refers to an evaluation of the customer intimacy components considering the customer
organization as a whole. The customer organization can be a team, a
business unit, or the entire enterprise (see chapter 1, figure 1.1).
As developed in chapter 2, achieving customer intimacy does not
solely consist of developing qualitative relationships with customers.
Customer intimacy relates to the management of business relationships as well as to the management of customer related knowledge.
More specifically, a successful customer intimacy strategy transforms
these relationships and knowledge into competitive advantages. The
decomposition of the concept of customer intimacy which is performed in this thesis in grounded on this analysis and roots in the
original definition of customer intimacy presented by Treacy & Wiersema (1993, p.87): “to continually tailor and shape products and services to fit an increasingly fine definition of the customer.” This definition can be split in two different parts: acquired customer intimacy
and leveraged customer intimacy:
• The acquired customer intimacy refers to obtaining and understanding this “fine definition of the customer.” It relates
to establishing business relationships and obtaining customer
related knowledge in order to determine means to adapt the
value proposition to the specific needs of each customer.
• The leveraged customer intimacy concerns the actual competitive advantages achieved through business relationships and
customer related knowledge. It represents the active part of
the customer intimacy definition: “to tailor and shape products and services”. These competitive advantages, such as customization and proactiveness are developed in section 4.4.

98

4. Customer Intimacy Breakdown Analysis

Leveraged Customer
Intimacy
Undirected
Adaptability

Customer
Intimacy

Standard
Solution for
Anonymous
Markets

Inflexible
Response to
Customer
Needs

Acquired Customer
Intimacy

Figure 4.1.: The Two Dimensions of Customer Intimacy
As illustrated in figure 4.1, both the acquired and leveraged customer
intimacy are required in order to effectively achieve a customer intimacy strategy:
• Considering the lower-left element of the quadrant which is defined as standard solution for anonymous markets, if the provider
does not manage customer related knowledge and business relationships in order to obtain information on the specific customer requirements, nor try to individually adapt its solution to
its customers, then this firm does not pursue customer intimacy
by any means and should try to become a product leadership
or operational excellence driven organization.
• The lower-right element – inflexible response to customer needs
– describes companies that have established business relationships and effectively gathered customer related knowledge. These organizations, however, are unable to put these into actions
in order to achieve a competitive advantage. For instance, if
a relationship manager presents customer requirements to the
provider organization, but the product development team rejects them and let the customer work with the standard offe-

4.2. Overview of the Customer Intimacy Breakdown Analysis

99

ring, then the customer is left out with an inflexible response
to its needs. This notion of “action on knowledge and relationships” reflects in part the definition of market orientation
presented in section 2.3.2: “the organization-wide responsiveness to the generation and dissemination of market intelligence
pertaining to current and future customer needs” (Jaworski &
Kohli, 1993, p.54). Thus, in this configuration, the provider does
not achieve a customer intimacy strategy with his customers.
• The upper-left element of this figure, called undirected adaptability, may be unrealistic. It refers to organizations which are
not aware of this “fine definition of the customer”: they do
not have knowledge about the needs and expectations of their
customers, nor business relationships to allow them to access
this information. However, they build their value proposition
around the creation of individualized solutions. Consequently,
their offering can only by chance fit their customers’ requirements and they also do not achieve customer intimacy.
• Finally, the upper-right element, which represents the actual
customer intimacy strategy, refers to organizations which both
obtain the fine definition of the customer and use it in order
to generate a competitive advantage: they acquire a certain degree of customer intimacy with their customers and they are
able to leverage it. Such organizations effectively manage both
customer related knowledge and customer relationships. They
also convert these two assets in a way that allows them to improve their value proposition and differentiate it from the standard ones proposed by their competitors.
In order to detail the requirements on organizations implementing
a customer intimacy strategy, the sections 4.3 and 4.4 further break
down the acquired and leveraged customer intimacy parts in multiple customer intimacy components. An overview of these components is proposed in figure 4.2.

100

4. Customer Intimacy Breakdown Analysis

Leveraged Customer Intimacy

Acquired Customer Intimacy
Acquired
Knowledge
(Individuals)

Acquired
Knowledge
(Organization)

Established
Relationships
(Individuals)

Established
Relationships
(Organization)

Customization

Customer Loyalty

Proactiveness

Cross-selling

Customer
Participation

Transaction Costs
Reduction

Figure 4.2.: Breakdown Analysis of the Acquired and Leveraged
Customer Intimacy

4.3. Acquired Customer Intimacy Components
The concept of acquired customer intimacy has been created in this
thesis in order to encompass the notion of “fine definition of the
customer” presented in the previous section. As introduced in chapter 2, customer intimacy differentiates itself from other marketing
strategies in the sense that it focuses on both customer related knowledge and customer relationships (see table 2.1). Customer related
knowledge is required in order to understand the customer’s current and future needs, as well as to determine which knowledge
should be provided to the customer. Then, establishing qualitative
customer relationships is necessary for a customer-intimacy driven
organization as relationships are the means to become a reliable and
trusted partner of the customer as well as to obtain further valuable
knowledge and information from the customer that can be used to
improve the value proposition. Manasco (2000, p.66) confirms that
“relationships and knowledge are inseparable” in order to capitalize
customer knowledge. Consequently, the two components pertaining
to the acquired customer intimacy are:
• Acquired customer knowledge
• Established customer relationships
As previously explained, a requirement in the approach followed
by this thesis is to perform the customer intimacy assessment at two

4.3. Acquired Customer Intimacy Components

101

different levels of analysis: the individual and the organizational levels. Indeed, in order to accurately understand the customer, it is necessary to identify the provider employees who have knowledge of,
and relationships with, the overall organization, as well as those who
have knowledge of, and relationships with, specific employees inside
the customer organization. Therefore, in this thesis, the two customer
intimacy components acquired customer knowledge and established
customer relationships are assessed at both the individual and organizational levels, as depicted in figure 4.2. These components are
further detailed in the next two parts of this section.

4.3.1. Acquired Customer Knowledge
The first component of the acquired customer intimacy refers to the
acquisition and development of customer knowledge. Batt (2004,
p.172) explains that in order to achieve customer intimacy, “the firm
must keep deepening its knowledge of the customer and put this
knowledge to work through the organization.” As a matter of fact,
customer intimacy requires advanced knowledge management capabilities. Zack et al. (2009) confirms that the organizations that pursue the value discipline customer intimacy have implemented the
widest range of knowledge management practices. Moreover, a positive correlation has been established between customer knowledge
development and service activities as “service relationships offer an
opportunity for greater customer knowledge to be developed by the
employees because of their repeated interactions with the same customer” (Gwinner et al., 2005, p.136). In a product development context, customer knowledge development has been defined as “a process of developing an understanding of customer new product preferences that unfolds through the iteration of probing and learning activities” (Joshi & Sharma, 2004, p.48). Taking a broader perspective,
Bueren et al. (2004) distinguish three categories of customer knowledge: about, for and from the customer.
Knowledge about the customer is certainly the most important one to
develop a customer intimacy strategy.

102

4. Customer Intimacy Breakdown Analysis

• At the organizational level, knowledge about the customer refers
to gaining an understanding of the current and future needs
of the customer, to obtaining information about the customer
strategy and about its mid- and long-term development. It also
includes knowledge about the interaction history with the customers such as the projects performed with the customer and
the products and services purchased by the customer. Knowledge about the customer also consists of the inherent description
of the customer organization which provides valuable information to optimize the interaction with the customer, such as
the organizational structure, the customer’s behavior and its
purchasing process. While some of this knowledge is certainly
explicit, such as the description of prior projects, opportunities
and contracts, a part of this knowledge is also implicit. For
instance, a project manager who completed successful projects
with the customer most likely gathered information about its
future needs and planned developments while a key account
manager is aware of its customer’s purchasing processes.
• At the individual level, knowledge about the customer refers to
knowledge about customer employees, such as specific needs,
preferences, and behavior. This aspect is particularly important
in a B2B context as the customer consists of multiple stakeholders, such as the users and buyers, which all have different requirements. In order to be successful and to optimize
its value proposition, the provider must be able to manage all
these different expectations (Homburg & Jensen, 2004). Gibbert
et al. (2002, p.3) argue that “smart companies [...] seek knowledge through direct interaction with customers, in addition to
seeking knowledge about customers from their sales representatives”. For instance, some customer employees may need a
specific service level agreement because they use a service differently from the rest of the organization.
Knowledge for the customer aims at fulfilling the customer’s needs
with regard to his knowledge requirements. It refers essentially to
information about the value proposition such as technical details on
the purchased products and services. This category of knowledge

4.3. Acquired Customer Intimacy Components

103

also includes insight into the customer’s industry which might be
relevant for the customer in order to generate future needs in the
customer organization such as new regulations, or new market opportunities. The consideration of multiple level of granularity, from
the entire organization perspective, down to the teams and the individuals perspective is also required as different customer teams
and customer employees will have different requirements in terms
of knowledge: depending on their role, they will expect business,
technical, or financial information.
Finally, knowledge from the customer consists of the information related to the products and services of the provider that the customer
employees acquire by using them. This includes information such
as the quality, reliability, or usability of the products and services.
This knowledge also includes information on the satisfaction of the
customer as well as suggestions from the customer for new products or service developments. If provider employees are able to access this knowledge and to convey it back in their organization, this
knowledge from the customer becomes a highly relevant asset for
adapting and improving the value proposition.

4.3.2. Established Customer Relationships
The second component of the acquired customer intimacy consists of
the relationships established between the provider and the customer,
at both individual and organizational levels. Relationships are an
inherent part of any business ecosystem and become steadily more
intensive in the current globalized economy (Donaldson & O’Toole,
2007). They have become an increasingly important matter of study
in marketing literature, as several analyses demonstrate their positive
influence on business performance (Narver & Slater, 1990; Varadarajan & Rajaratnam, 1986; Reichheld & Sasser, 1990).1 It has been explained in chapter 2 that the value discipline customer intimacy is
grounded in the concept of relationship marketing and relies on
the establishment of business relationships. In short, Donaldson
1

Further details are provided in section 2.2.1.2.

104

4. Customer Intimacy Breakdown Analysis

& O’Toole (2007, p.13) summarize the benefits derived from established business relationships in order to support a customer intimacy driven strategy: business relationships help “identifying customer needs and requirements, anticipating future trends and monitoring environmental forces, and satisfying customers’ existing and
future requirements.”
Business relationships have been assessed in multiple ways over the
past decades, and several studies which evaluate the constituents of
a relationship in a commercial setting are already available (Morgan
& Hunt, 1994; Odekerken-Schröder et al., 2003; Bove & Johnson, 2001;
Barnes, 1997). In previous literature, the assessment of customer relationship is referred to as relationship quality or relationship strength.
Even though Richard (2008) argues that much literature uses these
two terms equally, Bove & Johnson (2001, p.190, p.193) propose to
distinguish the two concepts. They define relationship quality as
“an overall construct which is based on all previous experiences and
impressions the customer has had with the service provider”, and relationship strength “as the magnitude of a relationship between two
individuals in a commercial setting.” In this perspective, relationship
quality is more focused on the organizational level while relationship strength concerns predominantly the individual level. In past
literature, the most often cited characteristics of relationship quality and relationship strength are trust and commitment2 (Richards
& Jones, 2008; Roberts et al., 2003; Lages et al., 2005). Therefore, assessing established relationships refers to understanding the degrees
of trust and commitment established between the provider, the customer, and their respective employees:
• Trust has been conceptualized as having “confidence in an exchange partner’s reliability and integrity” (Morgan & Hunt,
1994, p.23). It was further refined along the following three
dimensions: contractual trust, goodwill trust, and competence
trust (Sako, 1992). Contractual trust is determined by the re2

Communication quality, customer satisfaction, social bonds, and information
flows are further aspects that have been identified as characteristics of
relationship quality and strength.

4.3. Acquired Customer Intimacy Components

105

spective legal obligations of both partners. Goodwill trust refers
to a mutual commitment and support to each other, including confidence that the partners will not try to take an unfair
advantage of each other. Finally, competence trust has been
defined as the belief that the partner has the ability, technical
knowledge, expertise, and capability to perform his role (Sako,
1992).
• Commitment was defined by Anderson & Weitz (1992, p.19)
as “a desire to develop a stable relationship, a willingness to
make short-term sacrifices to maintain the relationship, and a
confidence in the stability of the relationship.” This translates
in the provider organization and its employees into a readiness
to help the customer solving his problems, into demonstrating
an adequate flexibility when needed by the customer, and into
seeking the best solution from the customer’s perspective on
the long-term rather than from the provider’s perspective on
the short term.
Considering the individual level of analysis, acquired customer knowledge and established customer relationships are intricately connected.
Ballantyne (2004, p.119) introduces the concept of relationship specific knowledge which he considers as a mediator for the development of trust and for the generation of business knowledge. He
argues that this is a “kind of tacit knowledge that might have positive use in dealing with current dilemmas and determining future
expectations.” Reciprocally, Gummesson (2008, p.190) establishes
the knowledge relationship as the 21st of his 30 “R” of relationship
marketing. He argues that knowledge is “not only embedded in
an individual, group, or corporation, but also in the relationships
between companies.” This knowledge relationship builds upon a
complex network of social ties established between provider and customer employees and it is referred to as a social structure (Donaldson
& O’Toole, 2007, p.116). This social structure, when used appropriately, is highly valuable for the provider as it can become a strategic
lever in order to improve the value proposition and to develop the
customer intimacy strategy (Dalkir, 2011, p.170). This potential value
of the social structure is called social capital. Dalkir (2011, p.474)

106

4. Customer Intimacy Breakdown Analysis

defines social capital has “the value created when a community or
society collaborates and cooperates (through such mechanisms as
networks) to achieve mutual values.” In this thesis, the assessment
of the established relationships at the individual level corresponds
to the assessment by means of social network analysis of the social
structure established between provider and customer employees.
In the next section of this chapter, the components pertaining to the
leveraged customer intimacy will be introduced.

4.4. Leveraged Customer Intimacy Components
The second part of the customer intimacy breakdown analysis is
called leveraged customer intimacy. While the acquired customer intimacy concerns the investments made by the provider in order to
obtain some knowledge of, and to establish some relationships with,
the customer, the leveraged customer intimacy refers to the actual
competitive advantages, benefits, and value proposition improvements achieved by the provider by “leveraging” these knowledge
and relationships. When the provider uses his knowledge of, and relationships with, the customer, he adapts, transforms, or enriches his
offering to the customer, thereby improving his value proposition,
and convincing the customer to choose him as a provider rather than
other competitors.
In order to fully understand the leveraged customer intimacy, a thorough review and analysis of literature has been performed in this
thesis. This analysis has led to decompose the leveraged customer intimacy into the following six components, as depicted in figure 4.2:
customization, customer loyalty, proactiveness, cross-selling, customer participation, and transaction cost reduction. The next parts
of this section elaborate on each of these components, outline their
association with the value discipline customer intimacy, and demonstrate why they lead to the generation of competitive advantages and
benefits for the provider. The actual metrics created in this thesis for
assessing these six components upon existing customer data will be
introduced in chapter 5.

4.4. Leveraged Customer Intimacy Components

107

4.4.1. Customization
Customization is the first component of the leveraged customer intimacy part of the model proposed by this thesis. Customer intimacy driven organizations, with their objective to “tailor and shape
products and services to fit an increasingly fine definition of the
customer” (Treacy & Wiersema, 1993, p.87) inherently rely on customization strategies which “aim at providing customers with individually tailored products and services” (Gwinner et al., 2005, p.131).
Customization is particularly important in the B2B context because
it is closely related to the servitization process that has occurred over
the past decades. Servitization, which refers to a business model shift
from selling products to selling “customer-focused combinations of
goods, services, support, self-service and knowledge” is, as a matter
of fact, a form of customization (Vandermerwe, 1988, p.314).
Several analyses in past literature have confirmed the importance of
customization in order to create a competitive advantage and to improve the value proposition. Fornell et al. (1996, p.8) demonstrated
with their American customer satisfaction index that customization,
which they defined as “the degree to which the firm’s offering is
customized to fit heterogeneous customer needs” has a more significant impact on customer satisfaction than reliability. Richards &
Jones (2008, p.126), in an analysis aiming at finding the value drivers
of customer relationship management, observed that “increased customization of products and services is positively related to brand
equity and relationship equity in the maintenance stage.” Thus, customization increases the provider’s value from the customer’s perspective. Finally, Vargo & Lusch (2004b, p.326) confirmed the importance of customization in contrast to standardization as they state
that “the normative marketing goal should be customization, rather
than standardization.” They thereby indicate that if standardization
increases production efficiency, it also decreases marketing effectiveness: the heterogeneity of the customer demand requires individually tailored response that standard offerings are unable to provide.
Therefore, organizations should consider customization rather than
standardization as their primary marketing focus.

108

4. Customer Intimacy Breakdown Analysis

In order to develop a customization strategy, different approaches
have been proposed, in particular, mass customization, customerization, and service customization through employee adaptiveness. An
analysis of these different concepts leads to the conclusion that customization, within the scope of this thesis, is aligned with the service
customization through employee adaptiveness approach:
• Mass Customization
Mass customization can be perceived as a means to combine
standardization with customization, thereby achieving both cost
efficiency and marketing effectiveness. It is defined as: “a system that uses information technology, flexible processes, and
organizational structures to deliver a wide range of products
and services that meet specific needs of individual customers
(often defined by a series of options), at a cost near that of massproduced items (Silveira et al., 2001, p.2). Mass customization
does not fit into the model proposed by this thesis as it leverages information technology rather than acquired knowledge
of, and established relationships with, customers in order to
achieve customization.
• Customerization
Customerization has been proposed as an evolution of mass
customization which gives more controls to customers in the
design of products and services, and relies on interactions with
customers to achieve customization (Wind & Rangaswamy, 2001).
Through an emphasis on knowledge from customers as well as
a redefinition of the role of the customer as an active co-creator,
customerization is, to some extent, close to customization as
proposed by this thesis. Customerization, however, does not
consider the development of interpersonal relationships with
the customer, but, similarly to mass customization, leverages
IT systems in which customers directly input their requests and
preferences in order to provide customers with customized solutions. Wind & Rangaswamy (2001, p.15) confirm that “mass
customization is IT-intensive on the production side, whereas
customerization is IT-intensive on the marketing side.” Thus,
customerization and customization in the context of this the-

4.4. Leveraged Customer Intimacy Components

109

sis are different concepts, even though some similarities can be
observed.
• Service Customization through Employee Adaptiveness
While mass customization and customerization intend to achieve
customization mainly though the use of information technology, Gwinner et al. (2005) emphasize the importance of the
provider employees in order to achieve customization in their
service customization through employee adaptiveness model. They
argue that customer knowledge is antecedent to effective customized service behaviors. However, in contrast to mass customization and customerization, customer knowledge is not
generated by information systems but resides in the front-line
employees who have regular interactions with the customer.
Thus, this model corresponds to the approach proposed by
thesis: the objective is to leverage the acquired knowledge of,
and the established relationships with, customers to thoroughly
understand the customers’ explicit and tacit needs. These assets are used to customize the offering and, thus, to achieve a
competitive advantage. Gwinner et al. (2005, p.136) confirm as
a matter of fact the importance of interpersonal relationships
for service customization as they argue that “service relationships offer an opportunity for greater customer knowledge to
be developed by the employees because of their repeated interactions with the same customer.”

4.4.2. Loyalty
Customer loyalty is the second leveraged customer intimacy component. In its definition of customer loyalty, Oliver (1999, p.34) insists
on the establishment of a stable and long-term relationship between
the provider and the customer: customer loyalty is “a deeply held
commitment to rebuy or repatronize a preferred product/service
consistently in the future, thereby causing repetitive same-brand or
same brand-set purchasing, despite situational influences and marketing efforts having the potential to cause switching behavior.” According to Treacy & Wiersema (1997, p.40), customer loyalty is the

110

4. Customer Intimacy Breakdown Analysis

most important benefit derived from a customer intimacy driven
strategy: “the customer-intimate company’s greatest asset is, not surprisingly, its customers’ loyalty.”

Multiple analyzes have corroborated the importance of customer loyalty over the past decades, since Reichheld & Sasser (1990) established that a 5% improvement in customer retention can lead to
a 25% to 85% profitability improvement. This finding created a
strong impulse for research that analyzes the relationship between
customer loyalty, customer retention, and customer satisfaction (Dick
& Basu, 1994). Reichheld & Teal (2001, p.39) elaborated five benefits
and competitive advantages which are derived from customer loyalty. These are the reduction of the customer acquisition costs, the
per-customer revenue growth, the operating costs reduction, the generation of referrals and recommendations, and the payment of price
premiums by loyal customers. Even though an empirical analysis in
the context of B2C financial services found that loyalty is not positively associated with profitability (Storbacka et al., 1994), Grönroos
(2007, p.8) and Heskett et al. (1994) confirmed that loyal customers
are in most cases profitable.

It has been widely recognized in past literature that well established
relationships are an antecedent to customer loyalty, thereby linking
customer loyalty to the acquired customer intimacy part of the model
proposed by this thesis. For instance, Hennig-Thurau et al. (2002)
established that the key aspects trust and commitments of relationship quality directly or indirectly impact the customer’s loyalty. Palmatier et al. (2007), in an analysis at both the organizational level
and at the employee level argued that relationship-enhancing activities, such as actions and efforts that strengthen relationship quality
positively influence both the loyalty to the sales persons and to the
provider organization. Finally, Ndubisi (2007) empirically proved
that relationship marketing endeavors are positively correlated with
an augmentation of the degree of customer loyalty.

4.4. Leveraged Customer Intimacy Components

111

4.4.3. Proactiveness
The breakdown analysis of the leveraged customer intimacy part of
this model has led to define proactiveness as the third leveraged customer intimacy component. With an emphasis on customers and
on customer needs, Sandberg (2007, p.253) defines customer-related
proactiveness as “acting based on the information gathered about the
customers before their behavior has had a direct impact on the firm,
or deliberately influencing and creating changes in customer behavior.” This definition outlines the importance of acquiring customerrelated knowledge and insight in the customer’s industry, and of
using this knowledge as a trigger of the customer related activities.
Thus, in a customer related proactiveness configuration, the provider
initiates the interaction process with the customer instead of awaiting its explicitly articulated demands. In a similar way, Treacy &
Wiersema (1997, p.127) confirm the importance of customer-related
proactiveness for successful customer-intimate organizations when
they argue that “a customer intimate firm uses its superior expertise in the client underlying problem to change the way the customer
does business.”
Proactiveness is often contrasted with reactiveness which indicates a
focus on understanding and fulfilling customer requirements, thereby
reacting to customer behavior (Sandberg, 2007). Thus, customerintimate organizations combine both reactiveness and proactiveness.
They are reactive as they work towards fulfilling to the highest extent
the customer needs. They are also proactive as they try to transform
and shape the customers operations, structures, and behavior in order to solve his problems, even before the customer is able to request
for it.
Different types of proactiveness have been investigated in previous
literature, in particular the proactive service improvement and the
proactive service recovery:
• Wallenburg (2009, p.78) focuses on proactive service improvement in a B2B context and brings proactiveness in the context
of innovation. Considering an innovation which is potentially

112

4. Customer Intimacy Breakdown Analysis

beneficial to the customer, either by leading to a cost reduction
or a performance improvement, a proactive improvements occurs if the provider “proactively enhances the service provided
to that specific customer” with this innovation without the customer asking for it. Wallenburg (2009) establishes that both
types of performance improvements are strong drivers of customer loyalty, thereby supporting the customer intimacy strategy.
• De Jong & De Ruyter (2004, p.458) elaborate on the importance
of adaptive and proactive behavior in service recovery. Adaptive behavior refers to the actions undertaken by employees in
response to specific customer problems whereas proactive behavior concerns problem-independent customer-related activities such as “soliciting suggestions from customers, detecting
and correcting causes of service problems and challenging existing routines.” De Jong & De Ruyter (2004) argue by means of
an empirical analysis that while adaptive behaviors positively
influence the customer’s degree of loyalty, proactive behaviors
lead to additional service revenues.

4.4.4. Cross-selling
The fourth leveraged customer intimacy component refers to the
cross-selling achievements of the provider. Kamakura et al. (1991)
explain that cross-selling aims at increasing the number of different
products and services sold to the customer and propose a predictive
model to assess the likelihood of the customer to accept cross-selling
driven offerings. Malms & Schmitz (2011, p.255) suggest a customer
intimacy aligned definition of cross-selling: “an offer of customized
solutions or the provision of a full assortment of products and services.” Reciprocally, taking the customer’s perspective, Venkatesan &
Kumar (2004, p.111) define cross-buying as “the number of different
product categories a customer has purchased.” They establish this
factor as a key element of their customer lifetime value assessment
model and prove that it increase the customer’s purchase frequency,
thereby generating additional revenues. In order to achieve crossselling, the provider should try to complement the original product

4.4. Leveraged Customer Intimacy Components

113

or service sold to the customer with other components that improve
the overall solution delivered to the customer.
Looking at the relationship between customer intimacy and crossselling, Akura & Srinivasan (2005, p.1008) demonstrate that customer
intimacy and cross-selling are intricately connected and argue that
firms “achieve customer intimacy when committing against a certain level of cross-selling.” Treacy & Wiersema (1997) confirm that
customer-intimate organizations inherently provide their customers
with cross-selling offering as they do not only sell products but solutions combining multiple products and services that fulfill the exact
customer’s needs. In the B2B context, Harding (2004) recognizes
the importance of cross-selling, but also argues that cross-selling can
damage the relationship if performed with the objective to increase
the provider’s revenues rather than to provide the customer with
the solution that fits its requirements and solves its problems. He
thereby links cross-selling with the component “acquired customer
knowledge” of the acquired customer intimacy part of this model
and confirms that deep customer knowledge is a prerequisite to effective cross-selling. This relationship between customer knowledge
and cross-selling has also been confirmed by Akura & Srinivasan
(2005, p.1007) who argue that “successful cross-selling requires customer intimacy and detailed information on customer preferences.”
Achieving cross-selling leads to multiple benefits for the provider. In
addition to the positive impact on revenues established by Venkatesan & Kumar (2004), cross-selling also improves customer’s profitability as the costs to acquire the customer can be distributed on
products and services of different categories. These costs are also reduced for any subsequent component added to the solution provided
to the customer. Cross-selling also has an indirect impact on the customer loyalty as it increases the customer’s switching costs and the
customer retention rates (Kamakura et al., 2003). If the customer purchased different products and services from the same provider, the
costs of replacing all these components by other alternatives is higher
than if he only bought one single product or service. Thus, an heterogeneous solution composed of multiple products and services is
a motivating factor for the customer to remain loyal to its provider.

114

4. Customer Intimacy Breakdown Analysis

Finally, it has also been established that cross-selling increases the
customer-related knowledge acquired by the provider (Kamakura
et al., 2003), thereby having a positive influence on acquired customer
intimacy. Indeed, the variety of products and services purchased by
the customer allows the provider to obtain a broader understanding
of the customer needs and preferences.

4.4.5. Customer Participation
Customer participation is the fifth component of the leveraged customer intimacy. It has been defined as “the customer behaviors related to specification and delivery of a service” (Cermak & File, 1994,
p.2). This aspect is fundamental in the previously described servicedominant logic which outlines that both the provider and the customer are co-creators: the customer is not a sole receiver of the value
distributed by the provider, but actively participates in its creation
by making his knowledge available to the provider (Vargo & Lusch,
2004a). Treacy & Wiersema (1997, p.136) confirm the importance of
customer participation for the success of a customer intimacy driven
strategy as they argue that “customer-intimate firms use their client
to stay at the edge of new thinking”. They quote an executive officer in a customer-intimate organization arguing that “the product is
conceived at the customer’s office” (Treacy & Wiersema, 1993, p.41).
Bettencourt (1997) identifies three different types of customer participation in his customer voluntary performance model:
• First, the customer can promote the provider organization and
its offering into its network. This kind of participation indicates, as previously explained, the degree of loyalty of the customer. However, it does not lead to a co-creation of the value
between the provider and the customer: the provider creates
the offering without the customer.
• Secondly, cooperation can be another form of customer participation: the customer supports the service employees to achieve
the expected service level agreements during the delivery phases,
but the knowledge of the customer is not used in order to support the design of the provided solution.

4.4. Leveraged Customer Intimacy Components

115

• The third type of customer participation is in line with the approach of this thesis and refers to customers who act as “organizational consultants”. Such customers actively participate in
the design and implementation of the solution by making available their understanding and knowledge of the problem to be
solved as well as by making recommendations for improving
the provided solution. In that regard, Bettencourt (1997) argue
that customers are a unique source of advice with an outstanding experience of the provider’s products and services.
Satzger & Neus (2010, p.230) emphasize the importance of customer
participation to support the provider’s innovations in their C4 framework. They suggest that customers are the most important source of
service innovation and argue that “the most efficient place for service
innovation may today lie outside of service provider organizations,
i.e. within peer-networks of users who are intrinsically motivated to
support innovation.” Similarly, Magnusson et al. (2003) empirically
compared innovations achieved by users and professional designers. Their finding was that users provided more original and userfocused innovations while professional provided innovation that are
easier to implement. Consequently, organizations having their customers participating to the value creation process hold an important
means to improve their value propositions and to achieve a competitive advantage.
Increasing customer participation in solution development also provides more structural benefits to the organization. Chesbrough (2007)
argues that open business models involving customers lead to a
reduction of the research and development costs and, thus, to an
increase in profitability. Customer participation also allows organizations to obtain qualitative market intelligence data and to better target the marketing strategy to customers and prospective customers (Ndubisi, 2007). Finally, Cermak & File (1994) established
that customer participation strengthens the relationship between the
customer and the provider as well as increases the customer satisfaction.

116

4. Customer Intimacy Breakdown Analysis

4.4.6. Transaction Costs Reduction
The sixth component of the leveraged customer intimacy refers to
a reduction of the transaction costs for organizations achieving customer intimacy. This notion of transaction costs has first been introduced by Coase (1937) as the costs of using the price mechanism
in the market. He established that the actual costs of the customer
for acquiring a product or a service in the market not only include
the price paid by the customer to the provider, but also information
costs, negotiation costs, and policy and enforcement costs. Similar
costs are borne by the provider for selling his product and services
since he has to inform the customer about its offering, negotiate the
offer, and ensure that the offering fulfills the customer’s expectations.
According to Dyer & Chu (2003), these additional costs are strongly
influenced by the existence of relationships and the establishment
of trust among the provider and the customer. For instance, the information costs are reduced if the customer chooses not to invest
time and resources to find a provider but simply select its preferred
partner. The negotiation will run more smoothly between partners
who already know each other. The customer’s enforcement costs
will be reduced if the customer trusts the provider in his ability to
deliver the expected product or service, as fewer safeguards have to
be setup. Dyer & Chu (2003, p.57) empirically proved that “trustworthiness lowers transaction costs and may be an important source of
competitive advantage.”
Williamson (1979) outlined the importance of transaction costs in the
study of economics and defined three aspects impacting transaction
costs: the transaction frequency, its uncertainty, and its idiosyncrasy,
which reflects the uniqueness and individualization of the investments performed by the provider and the customer, such as the
purchasing of special equipment by the provider in order to fulfill
the contract. Therefore, customer intimate organizations which make
some special efforts in term of time and cost investment in order to
fulfill the customer requirements increase the idiosyncratic degree of
the relationship and, thus, lower the transaction costs. Dyer (1997)
analyzed the influence of relationship-specific investments on trans-

4.4. Leveraged Customer Intimacy Components

117

action costs and confirmed that such investments do not lead to an
increase of the transaction costs and in some cases even lower them.
Focusing on the impact of customer loyalty, which is, as previously
described, customer-intimate organizations’ main asset, Reichheld &
Teal (2001, p.39) established that qualitative relationships with loyal
customers lead to a reduction of the acquisition and operating costs:3
• Several studies acknowledge that the acquisition of a new customer is significantly more expensive than the investments required to keep an existing customer (Grönroos, 2007, p.145).
Thus, by focusing on their most important and most loyal customers, customer intimate organizations have the means to lower
the acquisition costs. Treacy & Wiersema (1997, p.139) confirm
that customer intimate companies should avoid “business that
might generate only short-term revenues,” and whose acquisition costs cannot be balanced with regular and long-term revenues.
• Operating costs are reduced in long-lasting relationships with
loyal customers because the frequent and regular interactions
between the provider and the customer lead to the creation of a
common knowledge base between both organizations (Ballantyne, 2004). Thus, projects run in a smoother way as the provider better understands the customer’s expectations and the
customer can better articulate his requirements. In addition, in
the context of repeatedly delivered services, fewer mistakes occur as service are performed more often, which in turn leads to
an additional reduction of the operating costs (Grönroos, 2007,
p.146).
Zajac & Olsen (1993) and Den Butter (2010) consider the transaction
costs in the broader concept of value creation, and emphasize the notions of transaction value. Zajac & Olsen (1993) argue that the focus
on single party transaction cost optimization should be replaced by
a focus on transaction value and an emphasis on “joint value maxi3

Other economic effects established by Reichheld & Teal (2001) include
revenue growths, payment of price premiums, and referrals.

118

4. Customer Intimacy Breakdown Analysis

mization”. This transaction value takes into account the interdependencies between the provider and the customer. Den Butter (2010,
p.2) defines transaction management as “the ability to keep the costs
of trade transactions as low as possible so that the value creation
from these transactions is optimized.”
In summary, this chapter demonstrated by means of a thorough literature review that the concept of customer intimacy can be broken
down into two parts, namely the acquired customer intimacy and
the leveraged customer intimacy. The acquired customer intimacy
consists of the acquired customer knowledge and the established
customer relationships. The leveraged customer intimacy consists
of six components which are customization, loyalty, proactiveness,
cross-selling, customer participation, and transaction costs reduction. These customer intimacy components have been thoroughly described and their association to the concept of customer intimacy has
been motivated upon past literature. This analysis is foundational for
the remaining of this thesis. In chapter 5, it will be explained how
this thesis proposes to evaluate these components in an analytical
manner, thereby achieving the overall objective to assess the degree
of customer intimacy established with different customers as well as
its impact on business.

5. CI Analytics Model and
Methodology
In chapter 4, the concept of customer intimacy has been broken down
in multiple components which pertain either to the acquired or to
the leveraged customer intimacy. The objective of this chapter is to
introduce the CI Analytics model to assess and monitor these components as well as to detail the CI Analytics methodology to calibrate
and utilize the model. As it will be explained in the next sections,
these model and methodology use social network analysis and datamining techniques, and leverage customer related data available in
the provider’s information system. An essential benefit of this approach is that the assessment of the customer intimacy components
is performed automatically, once the calibration has been performed.
Thus, in line with business intelligence and analytics systems, this
thesis provides the ability to monitor the evolution of the customer
intimacy components values over time in a continuous manner.
Section 5.1 presents an overview of the CI Analytics model and methodology. Sections 5.2 and 5.3 subsequently detail the relevant sources
of customer intimacy data and the conceived metrics for assessing
the acquired and leveraged customer intimacy components.

120

5. CI Analytics Model and Methodology

5.1. CI Analytics Overview
While part 5.1.1 elaborates on the CI Analytics methodology, part 5.1.2
details the key aspects of the CI Analytics model.

5.1.1. CI Analytics Methodology
As described in chapters 2 and 4, the choice to follow the value discipline customer intimacy impacts and even determines the provider’s
strategy and operational model. A well-driven customer intimacy
strategy can be recognized along certain characteristics such as the
evolution of customer relationships into longer term partnerships,
the access to customers’ information systems, some regularity in the
interactions, the successful completion of joint activities with customers and the mutual involvement of top level management in these
activities. The provider’s information system contains elements of
evidence for most of these characteristics. For instance, successful
joint activities can be tracked in the project database. The interaction
regularity as well as the involvement of top-level management can be
assessed with an analysis of the different communication channels.
The development of a partnership can be derived from the information contained in the customer relationship management system.
The CI Analytics approach aims to identify the relevant elements of
evidence of the customer intimacy components inside the provider’s
information system as well as to define a means to aggregate them
into understandable customer intimacy metrics in order to assess the
degree of customer intimacy with each customer and at multiple levels of details. This approach poses three significant and interrelated
challenges which are solved by the CI Analytics methodology. These
three challenges are the following:
• Inference Challenge
The inference challenge concerns the fact that the customer intimacy components are not directly observable and measurable
inside the provider organization or at the interface between the
provider and the customer (De Choudhury et al., 2010). On the
contrary to physical characteristics such as size or volume, or

5.1. CI Analytics Overview

121

even to explicit performance metrics such as revenue or profitability, concepts such as established relationships or acquired
knowledge cannot be directly measured and, thus, must be inferred out of observable and available data, such as interactions,
projects, and revenue records.
• Relevance Challenge
The relevance problem relates to the fact that there is no exact
specification of the data which is necessary and how it should
be transformed in order to precisely infer each of the customer
intimacy components (De Choudhury et al., 2010). Indeed, the
available data inside the provider’s information systems can be
combined in an infinite number of customer intimacy metrics,
by simply varying the ways the different data items are aggregated. Thus, a key challenge is to identify the most relevant
sources of customer intimacy evidence as well as the best metrics which reflect the actual values of the customer intimacy
components. Moreover, a means to identify the actual values
of the different customer intimacy components must be determined in order to validate the proposed approach.
• Calibration Challenge
The third challenge makes the first two problems, the inference and the relevance issues, even more complex. This issue
roots in the fact that each provider organization has its own
way of interacting with its customers and manages customer
related data in a specific manner. For instance, some providers
prefer email communication while others prefer phone calls or
face to face meetings. A three-months project may be considered as long in some organizations and as short in others.
An organization may save all details about all interactions and
activities with a customer, while another one keeps only the
most relevant data. Thus, some metrics which are relevant for a
specific organization might become less significant for another
provider. Consequently, the generic customer intimacy metrics
which have been conceived must be adapted and weighted by
means of a calibration to the individual characteristics of each

122

5. CI Analytics Model and Methodology

provider organization, such as the interaction, activity, and data
storage patterns.
1

Break down the concept of customer intimacy
into mulitple components

Customer Intimacy
Breakdown Analysis

2

Identify the sources of evidence
to assess the customer intimacy components

Customer Intimacy
Evidence Sources

Define the customer intimacy metrics

Unweighted Set of
Metrics

3

4

Individually performed
for each provider

Calculate the customer intimacy metrics
5

6

Empirically estimate the
customer intimacy components
Calibrate the model by applying
data-mining techniques

7
Validate and interpret the model

Database Results
(Analytical)
Empirical Results
(Questionnaire)
Machine-Learning
Models
Weighted Set of
Metrics

Figure 5.1.: CI Analytics Methodology
The CI Analytics methodology intends to solve the inference, relevance, and calibration challenges, thereby adapting the model to the
specific data and interaction patterns of each provider. Since this
methodology relies on the analysis of customer related data in the
provider’s information system, its design is aligned to the knowledge
discovery in databases process proposed by Fayyad et al. (1996b) and
presented in section 3.2.1. The seven steps of the CI Analytics methodology are depicted in figure 5.1. They are detailed and put in relationship with the steps of the knowledge discovery in databases process in the next paragraphs. While the first three steps are generic
and performed once, steps 4 to 7 aim at solving the calibration challenge and, thus, are individually performed by each provider.

5.1. CI Analytics Overview

123

1. Break down the concept of customer intimacy into customer
intimacy components
The first step of the CI Analytics methodology refers to a thorough analysis of the concept of customer intimacy and its breakdown analysis into multiple assessable components. This analysis represents an important contribution of this thesis and is
elaborated in chapter 4. It establishes that customer intimacy
can be broken down into two parts, the acquired and the leveraged customer intimacy. The acquired customer intimacy consists of two components: acquired customer knowledge and
established customer relationships. The leveraged customer
intimacy consists of six components which are customization,
customer loyalty, proactiveness, cross-selling, customer participation and transaction cost reduction.
2. Identify the sources of evidence to assess the customer intimacy components
The next step of the CI Analytics methodology is concerned
with the identification of the relevant sources of evidence which
can be analyzed to infer the customer intimacy components. A
fundamental idea of the approach followed by this thesis is that
the degree of customer intimacy established between a provider
and customer is reflected to some extent in the provider’s information systems. In order to determine these relevant sources
of customer intimacy evidence, this thesis relies on previous
research and past literature in the field of relationship marketing, customer relationship management, and social network
analysis. The layer Customer Intimacy Data of the CI Analytics
model presented in figure 5.2 outlines the multiple sources of
customer intimacy evidence considered in the scope of this thesis.
3. Define the customer intimacy metrics to calculate customer
intimacy components out of the customer intimacy data
Closely related to the second step, the third step of this methodology consists of the actual design of the metrics which are
used to calculate the customer intimacy components out of
the available data in the provider’s information system. Pre-

124

5. CI Analytics Model and Methodology

vious literature provides numerous meaningful indications to
perform this activity. For instance, the Industrial and Marketing Purchasing Group (IMP) presented several contributions in
which they recognize that relationships are based on organized
patterns of interactions (Hakansson & Snehota, 2000, p.75). The
identification of these patterns is based on specific characteristics such as the quantity, intensity, and regularity of the interactions. These aspects are, thus, potential customer intimacy
metrics for assessing the relationships established between the
provider and the customer. In this step of the methodology,
the customer intimacy metrics are defined in a generic manner, and the most relevant metrics as well are their respective
weights are still unknown.
4. Calculate the customer intimacy metrics
In order to identify which of the generic customer intimacy
metrics are most relevant, the next step consists of calculating
them at both the organizational and individual levels. This activity corresponds to the data selection task in the knowledge
discovery in databases process presented in section 3.2.1. In
order to perform this calculation, the software CI Analytics has
been conceived and implemented in the scope of this thesis.
This software retrieves and transforms the available customer
data, calculates the customer intimacy metrics, and provides
the means to visualize their values. In its current version, this
application focuses on data which is available in the customer
relationship management system CAS genesisWorld.1 Further
details of the software CI Analytics are provided in chapter 6.
5. Empirically estimate the customer intimacy components
Similarly to step 4, this activity also corresponds to the data
selection part of the knowledge discovery in databases process.
To calibrate the CI Analytics model to the individual characteristics of a provider organization, some reference values for each
of the customer intimacy components are required. Indeed, this
methodology follows the supervised learning approach pre1

See http://crm.cas-software.com/EN/home.asp (accessed on 11.11.2011).

5.1. CI Analytics Overview

125

sented in section 3.2: the relevance of the customer intimacy
metrics which are calculated in step 4 is determined upon some
specific target values. The CI Analytics methodology proposes
to determine these reference values by means of a survey performed with the provider employees. Consequently, a questionnaire enabling the provider employees to estimate the acquired customer intimacy components has been designed. This
questionnaire contains multiple items which are presented in
section 5.2.4. Widely used Likert-type scales are used to measure the agreement or disagreement of the respondents to each
item (Miller & Salkind, 2002, p.330). While further details on
the design of the questionnaire are introduced in section 3.1.3,
a description of the actual survey performed with employees of
CAS Software AG to validate the methodology is provided in
chapter 7.
6. Calibrate the model by applying data-mining techniques
The step 6 of the methodology refers to the actual calibration
of the CI Analytics model. The aim of the calibration is to determine a means to combine the customer intimacy metrics calculated in step 4 in a way that this combination reflects the
reference values of the customer intimacy components which
have been empirically estimated upon a survey in step 5. In order to perform this task, data-mining techniques which aim at
discovering patterns in data sets are applied. Thus, this activity
corresponds to the steps pre-processing, transformation, and
data-mining of the knowledge discovery in databases process.
As explained in section 3.2.2, the machine learning algorithms
C4.5, support vector machine, k-nearest neighbor, and multilayer perceptron neural network are considered in the scope
of this thesis. Chapter 7 illustrates how this calibration is performed in a real scenario.
7. Validate and interpret the model
The last activity of the CI Analytics methodology refers to the
validation of the calibrated model and relates to the evaluation and interpretation tasks of the knowledge discovery in
databases process. It is necessary to assess the generalization

126

5. CI Analytics Model and Methodology

error of the proposed machine learning models, as well as to
confirm that the customer intimacy metrics can be used to assess the customer intimacy components. This evaluation is performed by means of a 10-times 10-fold cross-validation with
the performance indicators described in section 3.2.3. Finally,
the created machine learning models are interpreted in order
to derive some meanings, such as operational and managerial
implications out of the proposed calculation of the customer
intimacy components values.
The main aspects of the CI Analytics model such as the customer intimacy components, the customer intimacy metrics and the customer
intimacy data, have been introduced along the description of the CI
Analytics methodology. The next section summarizes these different
components and highlights their relationships.

5.1.2. CI Analytics Model
The diagram depicted in figure 5.2 illustrates the CI Analytics model.
This model consists of three main layers:
1. Customer Intimacy Layer
The first layer, called Customer Intimacy Layer, reflects the results of the breakdown analysis of the concept of customer intimacy in meaningful customer intimacy components based on
a literature review. This layer is the outcome of the first step of
the CI Analytics methodology. This breakdown analysis and the
resulting customer intimacy components are detailed in chapter 4.
2. Customer Intimacy Network
The second layer, defined as Customer Intimacy Network, consists of the different customer intimacy metrics which have
been designed in order to infer the customer intimacy components. To support the objective to provide an assessment
of the customer intimacy components with multiple levels of
details, a social network is used for representing the information contained in this layer. As explained in section 3.1.3, in

5.1. CI Analytics Overview

127

Layer 1: Customer Intimacy

Acquired Customer Intimacy

Leveraged Customer Intimacy

Acquired
Knowledge
(Individuals)

Acquired
Knowledge
(Organization)

Customization

Customer Loyalty

Proactiveness

Cross-Selling

Established
Relationships
(Individuals)

Established
Relationships
(Organization)

Customer
Participation

Transaction Costs
reduction
Based on
metrics

Layer 2: Customer Intimacy Network
Service Provider

Customer

Metrics

16

B

E
21

10

20

34

A
21

Interactions

19

Activities
C

F
32

Results

32

Centralities

G

D

Based on
customer
data

Layer 3: Customer Intimacy Data

Customer Interaction Channels

Customer Information Sources

Emails

Phone Calls

Project Database

CRM Application

Meetings

Letters

Support System

Other

Figure 5.2.: CI Analytics Model

128

5. CI Analytics Model and Methodology

this customer intimacy network, the vertices represent the provider and customer employees, and the edges and their respective weights are derived from the different customer intimacy
metrics. Moreover, this network representation provides the
ability to leverage specific graph based metrics called centrality
metrics.2 These centrality metrics provide a means to aggregate the metrics calculated at the individual level along multiple employees which form a team or a department. Three
main types of metrics have been identified and will be further
described in sections 5.2 and 5.3:
• Interaction metrics focus on the characteristics of the dialog
and exchanged informations between the provider and the
customer such as interaction regularity, quantity, or intensity.
• Activity metrics measure the efforts performed by the provider for the customer, such as the time spent to customize
a solution for the customer.
• Result metrics focus on the concrete achievements with the
different customers, such as sales and projects based metrics.
3. Customer Intimacy Data
The layer Customer Intimacy Data holds the underlying raw
data, the “evidence of customer intimacy”. Two main types of
sources can be distinguished:
• Customer interaction channels consist of the different means
used by the provider and the customer in order to exchange information, to dialog, and to jointly perform activities, such as emails, phone calls, letters, and face to face
meetings. As it will be explained in section 5.2, much literature confirms the close association between knowledge,
relationships, and interactions (Donaldson & O’Toole, 2007;
Gummesson, 2008; Hakansson et al., 2009).
2

Further details on centrality metrics are provided in chapter 3.

5.2. Assessment of the Acquired Customer Intimacy

129

• Customer information sources contain additional relevant data
for the calculation of the customer intimacy components
such as the project databases, the customer relationship
management system, or the support system storing the requests and issues of the customers.
To summarize the CI Analytics model, the customer intimacy components which are specified in the first layer (Customer Intimacy Layer)
result from the first step of the CI Analytics methodology, which is
the breakdown analysis of the concept of customer intimacy. The
values of these components are inferred from the metrics proposed
in the second layer (Customer Intimacy Network). These metrics are
calculated upon existing data which is available in the third layer of
the model (Customer Intimacy Data). The identification of this data
results from the step 2 of the CI Analytics methodology. The step 3
of the CI Analytics methodology provides a generic form of the customer intimacy metrics. The remaining steps 4 to 7 of the methodology enable a calibration of the proposed metrics to the specific
patterns of each provider.
In the next two sections, the steps 2 and 3 of the CI Analytics methodology are detailed for the acquired and for the leveraged customer
intimacy: the investigated sources of customer intimacy data as well
as the customer intimacy metrics designed to infer the acquired and
leveraged customer intimacy components are introduced.

5.2. Assessment of the Acquired Customer
Intimacy
It has been established in chapter 4 that the acquired customer intimacy can be broken down into two main components: acquired customer knowledge and established customer relationships. It has also
been determined that these two components should be assessed with
two levels of analysis: the individual and the organizational levels.
The individual level of analysis focuses on the acquired customer intimacy established by provider employees with customer employees

130

5. CI Analytics Model and Methodology

whereas the organizational level focuses on the acquired customer
intimacy established between provider employees with customer organizations.
In this section, part 5.2.1 is concerned with the identification of data
sources which are relevant for assessing acquired customer knowledge and established customer relationships. Part 5.2.2 and 5.2.3
focus on the actual metrics proposed by this thesis to calculate the
values of these two components at the individual and organizational
levels. Finally, part 5.2.4 elaborates on the series of Likert-items chosen to empirically assess them.

5.2.1. Using Interactions and Networks to Assess
Acquired Customer Intimacy
In the scope of this thesis, the main sources of customer intimacy
data for the assessment of the acquired customer intimacy components are the interaction and communication data. The positive correlation between relationships, knowledge, communication, and interaction has already been confirmed in numerous ways in past literature, as explained in the next paragraphs. Moreover, a key aspect of
this thesis lies in the application of network theory for the visualization and analysis of this assessment: a graph based representation is
used to depict the acquired customer intimacy at the individual level
and centrality metrics are calculated on these graphs to aggregate
the information up to the organizational level. Both Gummesson
and Batt confirm the relevance of this approach: while Gummesson
(1995) qualifies relationship marketing as “marketing seen as interactions, relationships and networks,” Batt (2004, p.171) states that
“interaction is the key construct at the heart of relationship marketing and the network paradigm.”
A major stream of research focusing on the analysis of business relationships through the study of interactions is driven by the Industrial and Marketing Purchasing Group (Hakansson & Snehota, 2000;
Leek et al., 2001).3 This group argues that relationships between buy3

See IMP website http://www.impgroup.org (accessed on 11.11.2011).

5.2. Assessment of the Acquired Customer Intimacy

131

ers and sellers are “built from interaction processes in which technical, social, and economic issues are dealt with” (Hakansson & Snehota, 2000, p.75). In this context, a business interaction is defined as
“the process that occurs between companies and which changes and
transforms aspects of the resources and activities of the companies
involved in it and the companies themselves” (Hakansson et al., 2009,
p.27). The underlying theory of the approach followed by the IMP
Group is the social / relational exchange theory which perceives relationships as social entities (Donaldson & O’Toole, 2007). From this
perspective, the actors are the provider and customer organizations
as well as the employees that belong to these organizations. These individuals create multiple interpersonal ties and are involved in multiple dialogs along the development of the interaction process. These
various aspects at different levels need to be considered in order to
understand the overall relationship (Medlin & Törnroos, 2006). The
emphasis is not on the provider side, but on the inter-organizational
relationships established between the provider and its customers at
both the organizational and individual levels: relationships belong
to a social structure and can be analyzed with social network analysis (Granovetter, 1985; Husted, 1994).
Three different research contributions have confirmed the potential
of investigating knowledge and relationships through an analysis of
social networks. First, Hutt & Walker (2006) have established a means
to determine the effectiveness of key account management programs
through an analysis of the centrality metrics related to the individual
key account managers. Then, focusing on the internal social network
established by employees inside a large consulting firm, Wu et al.
(2009) have determined metrics based on the topology of the social
network and correlated them with the success of teams and employees of this company. Finally, Kiss (2007) has developed a model
to leverage network data available in customer relationship management systems in order to improve customer classification tasks and
to support viral marketing activities.
Another important stream of research on relationships and interactions is conducted in the domain of service marketing by the “Nordic
School of Thought,” which looks at interactions as a marketing be-

132

5. CI Analytics Model and Methodology

havior (Grönroos, 2000). This perspective emphasizes that relationship marketing is effective if both the provider and the customer consider themselves as being part of a relationship: “a relationship has
developed when a customer perceives that a mutual way of thinking
exists between customer and supplier or service provider.” (Grönroos, 2007, p.36). The development of this relationship in a service
context is mainly supported by the interactions and communications
that occur between the provider and the customer. More precisely, a
two-way communication, a dialog, is essential for the development
of the relationship. Ballantyne (2004, p.114) argues that dialog, understood as an “interactive process of learning together”, provides
the means to obtain business knowledge as well as to develop trust
among the business partners.
Under the label of the “Nordic School of Thought”, Holmlund (2004)
developed a relationship framework in which the relationship is broken down into a flow of interactions in order to analyze its quality.
This flow contains three different levels of aggregation, as depicted
in figure 5.3:
• Act is the smallest element of the interaction flow and represents the “moment of truth” in the relationship. It consists of
an actual exchange of information or a joint activity, such as a
meeting, a phone call, or an email.
• Episode is a series of acts related to the same task, and forming
a minor part of the relationship, like a negotiation or a specific
part of a project.
• Sequence is a series of episodes and represents a major aspect of
the relationship like a whole project.
This model is relevant in the context of this thesis as it presents multiple benefits for the analysis of business relationships. According to
Holmlund (2004), it allows for a detailed analysis of the associations
between different relationship concepts, such as trust, commitment,
and value creation. It also provides the ability to compare different
relationships based on objective criteria, such as the duration of the
overall relationships or the number of sequences. Last, it provides a

5.2. Assessment of the Acquired Customer Intimacy

133

good framework for structuring empirical data when quantitative or
qualitative data is used.

Relationship
Sequence
Episode
A

A

A

Sequence

Episode
A

A

A

A

Episode
A

A

A

A

Sequence

Episode
A

A

Episode
A

A

A

Figure 5.3.: Interaction Levels in a Relationship (Holmlund, 2004)
The next section shows that the metrics created for the assessment of
the acquired customer intimacy components are inspired by the previously introduced network perspective and by the decomposition of
the relationship in series of sequences, episodes, and acts. The actual
interaction and communication data sources which are considered in
this thesis are the following:
• Meetings in persons, also called “face to face” meetings, in
which provider and customer employees exchange information
and knowledge and participate in joint activities;
• Phone calls, which are direct and synchronous communications
between providers and customers employees;
• Emails, as an asynchronous type of communication which may
involve multiple employees on both the provider and customer
sides;
• Letters, as another asynchronous communication channel.

5.2.2. Customer Intimacy Metrics at the Individual Level
The metrics at the individual level are those that assess to which extent a provider employee p knows a customer employee c and has
established a relationship with this person. One of the core objectives of this thesis is to provide an automated measurement of these

134

5. CI Analytics Model and Methodology

values based on the following data available in the provider’s information system: meetings, phone calls, emails, and letters. Two main
challenges reside in this activity:
• First, the confidentiality of the information must be respected:
for legal reasons it is not permitted to look into the content of
emails or letters, or to analyze the minutes of meetings and
phone calls. Therefore, the model proposed in this thesis focuses only on the “existence” of such data but not on the actual
content of the data itself. For instance, the information that several meetings between the provider and the customer occurred
over the past year can be used. However, the topics of these
meetings is not considered.
• The second challenge refers to the inference challenge described
in section 5.1. There is no means to know in advance which metrics should be calculated in order to assess the acquired customer intimacy components. For instance, the number of interactions, their duration, as well as the overall duration of the relationship are all potential indicators of these values. The problem is made even more complex as there is no available means
to weight the impact of the different interaction and communication channels. For instance, it cannot be argued that a certain amount of face to face meetings is more important than a
specific quantity of emails. In order to remedy this issue, the
concept of customer interaction time, which is explained in the
next paragraph, has been developed.
5.2.2.1. The Concepts of Customer Interaction Time and Weighted
Customer Interaction Time
Since it is not possible to estimate ex ante the relative importance
of the different customer interaction channels, this thesis proposes
to aggregate the overall time spent communicating and interacting
with the customer across all different channels in a value which is
called customer interaction time (CIT). In order to calculate this CIT
value, first, the different acts that belong to the relationships are evaluated in order to calculate their respective contribution to the overall

5.2. Assessment of the Acquired Customer Intimacy

135

CIT value. Then, these values are summed along each different interaction channel. Finally, the overall CIT value is aggregated as the
total customer interaction time across all interaction channels. For
instance, if a meeting lasting two hours is followed by two phone
calls which last respectively 10 and 20 minutes, the overall customer
interaction time is equal to 2 hours and 30 minutes. As opposed to
phone calls and meetings, the CIT value for emails and letters cannot be directly measured. Still, multiple functions can be defined and
calibrated, which take into account for instance the time to write or
read them. In this model, as a first approximation, we assume that
each email has a constant CIT value of demail and each letter has a
constant CIT value of dletter . Both demail and dletter can be individually
calibrated for each provider. In future research, these two constants
could be replaced by functions which take multiple parameters into
account such as the length of the emails and letters or the roles of the
senders and receivers.
An important parameter which determines the quality of the communication and interaction between two individuals is the number of
participants to the different interactions in which they are involved.
If a provider employee has a one-hour meeting with one single customer employee, he is more likely to obtain knowledge about this
person and to establish a relationship with this person than if he
meets this person in a larger event with several people involved. In
a similar way, if an email is sent by a provider employee to one person, it certainly contains more personalized information than if this
email is sent to all employees of the customer organization. Thus,
this model takes into account the number of participants to each interaction, and a second calculation of the customer interaction time
called weighted customer interaction time (wCIT) is provided. Further
details on the calculation of CIT and wCIT are presented in the next
paragraphs.
The CIT and wCIT values are not calculated only once for the overall
duration of the relationship, but can be evaluated for various time
intervals. This feature provides the ability to identify the multiple
episodes and sequences that belong to the relationship, as proposed
in the relationship framework presented in figure 5.3. Indeed, if no

136 Our idea is to aggregate the raw interaction
5. CI Analytics
Model
data with
the and Methodology
concept of episodes accross all communication channels

Sequence
Meetings
Emails

∆

Phone Calls

∆

Letters

Time
Dimension

Episode 1

Episode 2

Episode 3

Figure
5.4.: Customer Interaction Time: A Means To Aggregate CusResearch hypothesis: the regularity and intensity of episodes and sequences is an
tomer
Interaction
Across Multiple Channels
indicator of the
acquired
customer intimacy
•Ballantyne D. Dialogue and its role in the development of relationship specific knowledge. Journal of Business & Industrial Marketing.
2004;19(2):114-123.
•Gruner K. Does Customer Interaction Enhance New Product Success? Journal of Business Research. 2000;49(1)
•Yli-Renko H, Autio E, Sapienza HJ. Social capital, knowledge acquisition, and knowledge exploitation in young technology-based firms.
Strategic Management Journal. 2001;22(6-7):587-613.

High interaction, high knowledge
Regular interaction, better relationship

interaction occurred in any of the channels for a duration which is
above a certain interaction duration threshold ∆, this model assumes
that a new episode has begun. Figure 5.4 illustrates an example of
using the customer interaction time to identify the different episodes
and their respective compositions. In this example, the first episode
of the sequence consists of two meetings, three emails, three phone
calls and one letter. Then, no interaction occurs for a time period
which is equal to the interaction duration threshold ∆. This fact
indicates the beginning of the second episode which consists of six
emails, two phone calls and one letters, but no face to face meetings.
Finally, the third episode of the sequence starts after the interaction
duration threshold has been reached for a second time. It consists of
two meetings, four emails, two phone calls, and one letter.
16

28.09.2010

CIG Project: Measuring Customer Intimacy in B2B Services

Karlsruhe Service Research Institute
www.ksri.kit.edu

In order to define the customer intimacy metrics within this model,
further mathematical formalization is required. In this model, the
relationship is analyzed over a time period of duration T. This time
period T is divided in multiple contiguous time segments which all
have the same duration d. S = {s1 , ..., si , ..., sn } represents the set of
segments which compose the time period T. Thus, |S|, the cardinality
of S can be derived from the time period T and segment duration:
|S| = Td . A realistic example would be the interaction analysis over
the last year, defining the segment size as one month. In this case

5.2. Assessment of the Acquired Customer Intimacy

137

T = 12 months, d = 1 month, and |S| = 12. Another option would
be to increase the level of detail of the analysis and set the segment
size to one week. In this case |S| would be equal to 52.
p

p

CITc (si ) and wCITc (si ) represent the customer interaction time and
weighted customer interaction time between the provider employee
p and the customer employee c within the time segment si . They are
calculated as the sum of the customer interaction time (resp. weighted customer interaction time) of all interactions that occurred across
the four different channels between these two employees during si .
If H = {meetings, phonecalls, emails, letters} represents the set of interaction channels available to the provider employee p and the customer employee c, then:
CITpc (si ) =

c
( si )
∑ CITp,h

(5.1)

c
( si )
∑ wCITp,h

(5.2)

h∈ H

wCITpc (si ) =

h∈ H

The different components of the equations are calculated as follows:
c
c
• CITp,meetings
(si ) and wCITp,meetings
(si ) represent the total time
and total weighted time spent in meetings in which the provider p and c participated. If K cp,meetings (si ) symbolizes the set
of meetings within the time segment si in which both the provider employee p and the customer employee c participated, d j
the duration of the meeting j, and n j the number of participants
to the meeting j, without counting the employee p, then:
c
CITp,meetings
( si ) =

∑

dj,

(5.3)

dj
n
(s ) j

(5.4)

j∈K cp,meetings (si )
c
wCITp,meetings
( si ) =

∑

j∈K cp,meetings

i

138

5. CI Analytics Model and Methodology

c
c
• CITp,phonecalls
(si ) and wCITp,phonecalls
(si ) represent to the total time
and total weighted time spent in phone calls in which the provider p and c participated. If K cp,phonecalls (si ) symbolizes the set
of phone calls within the time segment si in which both the provider employee p and the customer employee c participated, d j
the duration of the phone call j, and n j the number of participants to the phone call j, without counting the employee p,
then:
c
CITp,phonecalls
( si ) =
dj
(5.5)
∑
j∈K cp,phonecalls (si )

∑

c
wCITp,phonecalls
( si ) =

j∈K cp,phonecalls

dj
n
(s ) j

(5.6)

i

c
c
• CITp,emails
(si ) and wCITp,emails
(si ) represent the importance of
c
email communication. If K p,emails (si ) symbolizes the set of emails
exchanged between p and c within the time segment si , and n j
the number of recipients of the email j, then:
c
CITp,emails
(si ) = |K cp,emails (si )|.demail

(5.7)

demail
nj
(s )

(5.8)

c
wCITp,emails
( si ) =

∑

j∈K cp,emails i

c
c
• CITp,letters
(si ) and wCITp,letters
(si ) represent the importance of
c
mail communication. If K p,letters (si ) symbolizes the set of letters
exchanged between p and c within the time segment si , and n j
the number of recipients of the letter j, then:
c
CITp,letters
(si ) = |K cp,letters (si )|.dletter

(5.9)

dletter
nj
(s )

(5.10)

c
wCITp,letters
( si ) =

∑

j∈K cp,letters

i

5.2. Assessment of the Acquired Customer Intimacy

139

In order to identify the interaction episodes between the provider
employee p and the customer employee c, the set of episodes within
the time period T is denoted EPpc . The principle for the identification
of the different episodes is as follows: the previously introduced interaction duration threshold ∆ is set as proportional to the segment
duration d: ∆ = λ × d, λ ∈ N. If no interaction occurs within contiguous segments whose total duration is equal to ∆, then the next
interaction indicates the beginning of a new episode. In this model,
∆ is equal to the segment duration d (λ = 1). Thus, if no interaction
occurs within one time segment, the next interaction indicates the beginning of a new episode. The length of the episode is, consequently,
determined by the number of contiguous segments in which some
interaction occurred.
A restriction in this model is that, so far, the actual CIT and wCIT values are not considered for the identification of the episodes. Even a
small interaction, like one email, is sufficient in order to have the segment it belongs to being considered as part of an episode. Therefore,
two additional threshold values, the interaction quantity threshold b
and the weighted interaction quantity threshold wb have been created. Using these parameters, a segment si is ignored in the identification of the episode if the corresponding customer interaction time
CIT and weighted customer interaction time wCIT are below their
respective threshold b and wb: CITpc (si ) < b and wCITpc (si ) < wb
Figure 5.5 illustrates the identification of the different episodes upon
the analysis of the segments. In this example, b and wb are both set
to zero and ∆ is set to the segment duration d. The first episode e1
is mapped to the first segment s1 . Then, no interaction occurs with
the segment s2 and, therefore, a new episode starts with the segment s3 . Both s3 and s4 contain multiple interactions. Therefore,
the second episode is spread over these two segments. Since the
segment s5 and s6 do not contain any interaction, the third episode
starts with the segment s7 . Consequently, using this approach, three
episodes can be identified over this time frame consisting of seven
segments. Importantly, the length of the analyzed time frame T as
well as the segment size d have a significant role in the identification

140

5. CI Analytics Model and Methodology

Meetings
Emails
Phonecalls
Letters
Time
Segment s1 Segment s2 Segment s3 Segment s4 Segment s5 Segment s6 Segment s7

Episode e1

Episode e2

Episode e3

Timeframe
T

Figure 5.5.: Segmentation of the Relationship to Identify Episodes
Across Multiple Channels
of the episodes and, thus, have to be carefully chosen at the calibration time.
5.2.2.2. Acquired Customer Intimacy Assessment at the
Individual Level
The metrics conceived in this thesis to assess the acquired customer
intimacy are based upon four interaction characteristics which have
already been identified in past literature: quantity, intensity, regularity, and mode of interaction. These characteristics will be further
developed in the next paragraphs. Based on the previously introduced concepts of segment, episode, customer interaction time, and
weighted customer interaction time, eight metrics which reflect these
four interaction characteristics have been created.
Since some of these metrics consider only the most relevant segments
within the time frame T, we define Scp as a subset of the previously
defined set S of segments which exclusively includes segments for
which CITpc and wCITpc are above the previously mentioned interac-

5.2. Assessment of the Acquired Customer Intimacy

141

tion quantity thresholds b and wb:
Scp = {si ∈ S | CITpc (si ) > b ∧ wCITpc (si ) > wb}

(5.11)

1. Quantity: volume and weighted volume
The first metrics, called volume and weighted volume, relate
to the quantity of interactions between the provider employee
p and the customer employee c. Interaction quantity has already been investigated as a potential indicator of relationships
in past literature (Nezlek, 2003): it is more likely that p has established a relationship with c and has acquired some knowledge
about c if p and c have a high volume of interaction rather than
a low interaction quantity. Volume (resp. weighted volume)
is calculated as the customer interaction time (resp. weighted
customer interaction time) between these two individuals along
the time frame T. The calculation of these two metrics is based
on the following equations:
Volumecp =

∑ CITpc (si )

(5.12)

∑ wCITpc (si )

(5.13)

si ∈ S

wVolumecp =

si ∈ S

2. Intensity and weighted intensity
The second metric refers to the intensity of the interaction. A
certain volume of interaction can be reached upon either multiple small and sporadic low-intensity acts, or with a limited
number of acts with a higher intensity. The influence of the
interaction intensity on knowledge flow and relationships has
also been studied in multiple ways in a business context. Noorderhaven & Harzing (2009, p.2) identify in their research that
“intensive social interaction provides opportunities for social
construction of knowledge in a learning dialogue”. Similarly,
Bennett & Robson (1999) argue that intense interactions support the exchange of information and is a means to overcome
the challenge of knowledge and information asymmetry. Finally, Hakansson et al. (2009, p.81) confirm that interaction in-

142

5. CI Analytics Model and Methodology

tensity influences the effects of the interaction on the involved
resources from both the provider and the supplier. The metrics intensity (Intensitycp ) and weighted intensity (wIntensitycp )
are calculated as the average customer interaction time (resp.
weighted customer interaction time) calculated over the segments which belong to Scp :
Intensitycp =
wIntensitycp

=

∑si ∈Scp CITpc (si )

|Scp |
∑si ∈Scp wCITpc (si )

|Scp |

(5.14)
(5.15)

3. Regularity: frequency, duration and number of episodes4
The third interaction characteristic considered in this model
concerns the regularity of the interactions. Regular communication and interaction have been recognized as a key aspect of
successful relationship marketing (Berry, 1995). Kong & Mayo
(1993) confirm the importance of the regularity dimension as
they argue that ‘successful business-to-business relationships
are based on regular, constructive and innovative interaction.”
Focusing on communication effectiveness in professional services, Sharma & Patterson (1999, p.163) consider that “regular communications can help develop a sense of closeness and
ease in the relationship, and be instrumental in building emotional and social bonds.” Therefore, three metrics have been
conceived in order to assess the regularity of the interactions:
frequency, duration, and number of episodes.
• Frequency refers to the proportion of segments in which
some interactions between the provider employee p and
the customer employee c happened within the time period
T. If p and c communicated and interacted in multiple
different segments, they are more likely to have a regular
4

The weighted version of these metrics are not necessary as their values
would be equal to their corresponding “non weighted” versions.

5.2. Assessment of the Acquired Customer Intimacy

143

interaction than if they only interacted within one or two
segments. Frequency (Frequencycp ) is calculated as the percentage of segments that belong to Scp to the total number
of segments within the time frame T:
Frequencycp

|Scp |
=
|S|

(5.16)

• Duration indicates to which extent the interactions between p and c span over the time frame T. A certain frequency value indicates the number of segments in which
some interactions occurred, but it does not specify whether
these segments are concentrated in a specific part of T,
such as the beginning or the end of T, or if they are uniformly distributed over T. This aspect, however, is significant in the interpretation of the interactions: if the segments are contiguous to each other, it means that p and c
had regular interactions over a limited period of time only.
On the opposite, if the interactions happened in the first
and last segments of the time period T, p and c certainly
had more regular interactions. In order to calculate the
metric duration, it is necessary to identify the index of the
first and last segments of time period T which contain relevant interactions. Considering the set of segments S and
its subset Scp both chronologically ordered, we define f as
the index in S of the first item in Scp and l as the index in
S of the last item in Scp . For instance, if the first and last
relevant interactions occurred respectively in the third and
seventh segments, then f =3 and l=7. Using these values,
the metric duration (Durationcp ) is calculated as follows:
Durationcp =

l− f +1
|S|

(5.17)

• Number of episodes is the last conceived metric as indicator of the interaction regularity. It is derived from the rela-

144

5. CI Analytics Model and Methodology

tionship framework presented in figure 5.3. Considering a
certain frequency value, a high number of episodes would
indicate several interruptions in the relationship while a
low number of episodes denotes a continuity in the interaction as several segments would be contiguous to each
other. The metric number of episodes (NumberEpisodescp )
is indicated by the cardinality of the previously introduced
set of episodes EP:
NumberEpisodescp = | EPpc |

(5.18)

4. Mode of interaction
The metric mode of interaction (Modecp ) indicates the proportion of face to face meetings in the overall interaction between
the provider employee p and the customer employee c. This
metric is derived from the finding that meetings in person have
a higher significance in the construction of the relationship and
in the exchange of knowledge than the other channels of interaction such as phone calls and emails. In an analysis of 35
sales and services virtual teams, Kirkman et al. (2004) identify
that the number of face to face meetings is a moderating factor between the teams’ empowerment and their performance.
Noorderhaven & Harzing (2009, p.2) confirm that “face-to-face
social interactions form a communication channel particularly
conducive to the transfer of tacit, non-codified knowledge.”
Mode of interaction is calculated with the following equation:
Modecp

c
( si )
∑si ∈S CITp,meetings
=
∑si ∈S CITpc (si )

(5.19)

5.2.2.3. A Graph-Based Representation of the Customer Intimacy
Metrics
The customer intimacy metrics defined in the previous paragraph
are the indicators of some specific interaction patterns between a
provider employee and a customer employee. Taking the broader

5.2. Assessment of the Acquired Customer Intimacy

145

perspective of the “many-to-many” interactions occurring between
all employees of the provider P and of the customer C (Gummesson,
2008), it is possible to use these metrics in order to design multiple
graph-based representations of the social network established between the two companies. In these graphs, the vertices are the provider and customer employees, the edges indicate some interactions
among these employees, and the weights of the edges are calculated
as a function of the previously defined customer intimacy metrics.
The set of graphs that can be inferred out of the customer intimacy
metrics is defined as G = { G1 , Gk , ..., Gn }. Gk is the graph that uses
the “weighting” function ωk , as explained in section 3.1.1, to calculate the weights of the edges. More formally, the graph Gk is defined
as Gk = (V, Ek ). V is the set of vertices of the graph Gk and consists of the employees involved in the interaction between P and C.
V is composed of the two subsets VC and VP where VP represents
the employees of the provider organization and VC the employees of
the customer organization: V = VP ∪ VC . If e p,c represents the edge
between the provider employee p and the customer employee c, and
w p,c the weight of the edge e p,c , then Ek represents the set of edges in
the graph Gk which link the provider and customer employees and
whose weights are calculated with the function ωk :
Ek = {e p,c ; w p,c = ωk (e p,c ) | ∀ p ∈ VP ; ∀c ∈ VC ; ωk : Ek → R+ } (5.20)
The customer intimacy metrics presented in the previous paragraphs
represent some standard functions for calculating the weights of the
edges. Indeed, the weights could be defined by the volume or by the
intensity of the interactions between the provider employee p and the
customer employee c. In these cases, the weighting function would
be respectively:
ωVol (e p,c ) = Volumecp
(5.21)
ω Int (e p,c ) = Intensitycp

(5.22)

Figure 5.6 illustrates the representations of the social network formed
by provider and customer employees using the metrics volume and
intensity as weighting functions. While the volume of interaction is

146

5. CI Analytics Model and Methodology

Provider
P P
Provider

Customer
C C
Customer
2 2

P1 P1
1 1

C1 C1

Provider
P P
Provider
P1 P1

2 2 C1 C1
1 1 0.5 0.5

4 4

P2 P2

P2 P2
2 2 C2 C2

P3 P3

Customer
C C
Customer

0.7 0.7 C2 C2

P3 P3

1 1
1 1

0.5 0.5
1 1

C3 C3
P4 P4

4 4

C3 C3
P4 P4

2 2

(a) weighting function based on volume (b) weighting function based on intensity

Figure 5.6.: Two Different Graph Representations of the Social Network Formed by the Provider and Customer Employees
used to calculate the weights of the edges in graph 5.6(a), the intensity is used in graph 5.6(b). The provider and customer employees involved in the interaction are: VP = { P1, P2, P3, P4} and VC =
{C1, C2, C3}. It can be observed that the weights of the edges are
significantly different on both graphs. For instance, in graph 5.6(a)
the edges e P3,C1 and e P4,C3 both indicates higher volumes of interaction than the other edges. However, in graph 5.6(b), it is shown that
the edge e P4,C3 has a high value of intensity but the edge e P3,C1 has
a low value of intensity. This example illustrates that the chosen
weighting function significantly impact the resulting representation
of the social network formed by the provider and customer employees.
This thesis aims at leveraging the conceived customer intimacy metrics in order to infer the values of the customer intimacy components, which are at the individual level, the acquired knowledge of,
and established relationships with, customer employees. The objective from a social network analysis perspective is, thus, to determine
the two weighting functions ωKnowledge and ω Relationship , so that the
weights of the edges in the graph representation indicate respectively

5.2. Assessment of the Acquired Customer Intimacy

147

the acquired knowledge of, and established relationships with, customer employees. In order to identify these weighting functions, the
steps 4 to 7 of the CI Analytics methodology are performed. These
steps are detailed in chapter 7.
The next section elaborates on the metrics conceived to assess the
acquired customer intimacy at the organizational level.

5.2.3. Customer Intimacy Metrics at the Organizational
Level
Focusing on the acquired customer intimacy part of the CI Analytics
model presented in figure 5.2, the metrics at the individual level indicate to which extent employees of the provider organization have
established relationships with, and acquired knowledge of, customer
employees. Similarly, as explained in chapter 4, the metrics at the organizational level indicate the relationship that a provider employee
has established with a specific customer organization as well as the
amount of knowledge he has acquired on this organization.
Two different types of metrics are proposed by this thesis to assess
the acquired customer intimacy at the organizational level:
• first, the concepts of customer interaction time and weighted
customer interaction time are adapted so that an assessment
of these values at the organizational level is applicable. Thus,
the metrics at the individual level presented in section 5.2.2 can
also be calculated at the organizational level.
• second, further metrics which leverage the characteristics of the
previously introduced social network established between the
provider and customer organizations are defined. These metrics are the degree centrality, the normalized degree centrality,
and the normalized closeness centrality.
5.2.3.1. The Concept of Customer Interaction Time at the
Organizational Level
The previously introduced customer interaction time CITpc1 (si ) and
weighted customer interaction time CITpc1 (si ) are calculated using

148

5. CI Analytics Model and Methodology

some specific aggregation functions of the interactions occurring between the provider employee p and the customer employee c1 across
all interaction channels within the time segment si . During the segment si , the provider employee p may not only have interacted with
the customer employee c1 , but also with the employees c2 and c3
from the same customer organization C. Consequently, it is possible
to calculate in the same manner the customer interaction time and
weighted customer interaction time between p and c2 and between
p and c3 , leading to the values CITpc2 (si ), wCITpc2 (si ), CITpc3 (si ) and
wCITpc3 (si ). By aggregating these multiple values, it is possible to
obtain the customer interaction time and weighted interaction time
between the provider employee p and the customer organization C
or any of its subsets such as the team C1 formed by the employees c1 ,
c2 , and c3 .
More formally, the set of employees of the customer organization C
is defined as VC . If Cx represents the subset x of the organization C,
such as a team, a department, or a business unit, VCx represents the
set of employees which belong to Cx . Thus VCx is either included in
VC or equal to VC : VCx ⊆ VC . It is then possible to calculate the CIT
and wCIT values between the provider employee p and Cx with the
following equations:
CITpCx (si ) =

∑

CITpc (si )

(5.23)

∑

wCITpc (si )

(5.24)

c∈VCx

wCITpCx (si ) =

c∈VCx

It is also possible to calculate the customer interaction time and
weighted customer interaction time for any specific interaction channel at the organizational level. As defined previously, if H = {meetings,
phonecalls, emails, letters} represents the set of interaction channels
available to the provider employee p for interacting with employees
of the customer organization C, then:
Cx
CITp,h
( si ) =

∑

c∈VCx

c
CITp,h
(si )|∀h ∈ H

(5.25)

5.2. Assessment of the Acquired Customer Intimacy

Cx
wCITp,h
( si ) =

∑

c
wCITp,h
(si )|∀h ∈ H

149

(5.26)

c∈VCx

5.2.3.2. Organizational Metrics Based On Customer Interaction Time
Using the concepts of customer interaction time CIT and weighted
customer interaction time wCIT applied at the organizational level,
it is possible to calculate the metrics presented in section 5.2.2 for any
organization C or subset of this organization Cx , such as a team, a
department or a business unit. Since these metrics are described and
motivated in detail in section 5.2.2, this section mainly outlines the
equation defined in order to adapt them to the organizational level.
In order to provide the ability to ignore the time segments in which
a certain level of interaction has not been reached for the calculation
of the customer intimacy metrics, the interaction quantity thresholds
B and wB are defined at the organizational level. SCx
p represents
the subset of time segments within the time period T for which the
overall customer interaction time for all employees that belong to the
organization Cx (resp. weighted customer interaction time) is above
the threshold B (resp. wB):
SCp x = {si ∈ S|CITpCx (si ) > B ∧ CITpCx (si ) > wB}

(5.27)

The customer intimacy metrics at the organizational level which leverage the concepts of customer interaction time and weighted customer
interaction time are based on the four following interaction characteristics: quantity, intensity, regularity, and mode of interaction.
1. Quantity: volume and weighted volume
The metrics volume (VolumeCp x ) and weighted volume (wVolumeCp x )
are indicators of the quantity of interaction that occurred between the provider employee p and the subset Cx of the organization C. These values are calculated as the sum of the customer interaction time (resp. weighted customer interaction
time) for all customer employees that belong to the organization subset Cx along all time segments si in the time period T:

150

5. CI Analytics Model and Methodology

VolumeCp x =

∑ CITpC (si )

(5.28)

∑ wCITpC (si )

(5.29)

x

si ∈ S

wVolumeCp x =

x

si ∈ S

2. Intensity and weighted intensity
The metrics intensity (IntensityCp x ) and weighted intensity (wIntensityCp x ) indicate whether the relationship with the customer employees that belong to Cx is based on multiple small interactions
or on fewer acts which have a longer duration. Their calculation, thus, is based on the average customer interaction time
(resp. weighted customer interaction time) along the relevant
time segments which belong to SCp x .
IntensityCp x
wIntensityCp x

=

=

∑si ∈SCp x CITpCx (si )

|SCp x |
∑si ∈SCp x wCITpCx (si )

|SCp x |

(5.30)

(5.31)

3. Regularity: frequency, duration, number of episodes
The regularity of the interaction indicates whether the interactions with the customer employees are evenly spread along the
different time segments or if they are concentrated in some specific segments, for instance at the beginning of at the end of the
time period T. In order to assess the interaction regularity, three
metrics have been defined: frequency, duration, and number of
episodes:
• Frequency (FrequencyCp x ) represents the proportion of time
segments in which relevant interactions occurred between
the provider employee p and employees of the customer
that belong to Cx within the time period T. It is calculated
as follows:
|Scp |
Cx
Frequency p =
(5.32)
|S|

5.2. Assessment of the Acquired Customer Intimacy

151

• Duration (DurationCp x ) describes to which extent the interactions between the employee p and employees of the customer organization Cx span over the time period T. If f
and l represent respectively the index of the first and last
segments which contain some relevant interactions between p and Cx , then the metric duration is calculated as
follows:
l− f +1
DurationCp x =
(5.33)
|S|
• Number of episodes (NumberEpisodesCp x ) relates to the continuity of the relationship, as a small number of episodes
indicates that the segments in which some relevant interaction occurred are contiguous to each other, while a higher
number of episodes indicate some interruptions between
the different interactions. If EPpCx represents the set of
episode between the provider employee p and the customer organization Cx , the metric number of episodes is
calculated as the cardinality of EPpCx :
NumberEpisodesCp x = | EPpCx |

(5.34)

4. Mode of interaction
The metric mode of interaction ModeCp x refers to the proportion
of time spent with meetings in person between the provider
employee p and the employees of the customer organization Cx
on the overall interaction time. It is calculated with the following equation:
C

ModeCp x

=

x
( si )
∑si ∈S CITp,meetings

C

∑si ∈S CITp x (si )

(5.35)

5.2.3.3. Organizational Metrics Based On Network Theory
In addition to the eight previously defined metrics which are based
on the concepts of customer interaction time and weighted customer

152

5. CI Analytics Model and Methodology

interaction time, it is also possible to leverage the graph based representations presented in section 5.2.2. The main advantage of these
additional metrics is that they take into account not only the customer employees with whom some interaction occurred, but also all
the remaining potential customer employees with whom no relationship so far have been established. These are, thus, further indicators
of the integration of the provider organization in the customer organization. For each graph representation Gk of the social network
established between the provider and customer organizations P and
C, it is possible to calculate the following metrics:
• Number of contacts (degree centrality)
The metric number of contacts (NumberContactsk ( p, Cx )) is calculated as the degree centrality of the provider employee p with
the customer organization Cx on the graph Gk . This metric is
presented in section 3.1.2. Using the metric number of contact,
it is possible to determine the number of customer employees
with whom a provider employee interacted over a specific time
period. This metric, thus, provides the ability to compare the
relationship established by different provider employees with
the customer organization Cx .
• Normalized degree centrality
The normalized degree centrality of a node i is presented in
section 3.1.2. It is defined as the number of edges incident to i
divided by the maximum number of potential nodes adjacent
to i. Within this model, the normalized degree centrality of
the provider employee p with the organization Cx on the graph
0k
Gk is denoted Cdegree
( p, Cx ). It indicates the proportion of individuals in the organization C with whom the employee p has
established some relationships. The normalized degree centrality enables the comparison of the relationship established by a
specific provider employee p with multiple customer organizations which all have a different number of employees.
• Normalized closeness centrality
The normalized closeness centrality is defined in section 3.1.2.
It has been established in past literature that closeness central-

5.2. Assessment of the Acquired Customer Intimacy

153

ity is an indicator of the effectiveness of the ability to communicate and convey information within a defined network (Freeman, 1979; Beauchamp, 1965). Within this model, the closeness
0k
centrality (Cclose
( p, Cx )) of the employee p with the organization
Cx on the graph Gk complements the metrics intensity and volume and indicates the proximity of the employee p with the
organization Cx .
In summary, eight metrics have been conceived to assess the acquired
customer intimacy at the individual level and 11 metrics have been
conceived to assess it at the organizational level. These metrics are
listed in table 5.1.
The next section elaborates on the series of Likert-items proposed
by this thesis in order to empirically assess the acquired customer
knowledge and established customer relationships.

5.2.4. Empirical Assessment of the Acquired Customer
Intimacy
As explained in section 5.1.1, the fifth step of the CI Analytics methodology requires to perform an empirical assessment of the customer
intimacy components acquired customer knowledge and established
customer relationships in order to complete the calibration of the customer intimacy metrics. To perform this assessment at both the individual and organizational levels, a series of items assessed on Likerttype scales has been conceived. Further information on Likert-type
scales is presented in section 3.1.3 and an illustrative questionnaire
is proposed in appendix A.
• Acquired Customer Knowledge
In order to ensure the relevance and validity of the items used
in this thesis to empirically assess acquired customer knowledge, these items are derived from the well recognized scales
created by Gwinner et al. (2005), Jayachandran et al. (2004), and
Joshi & Sharma (2004). Gwinner et al. (2005) assessed customer
knowledge with two items in order to determine the influence

154

5. CI Analytics Model and Methodology

Table 5.1.: Customer Intimacy Metrics at the Individual and Organizational Levels
Customer Intimacy Metric

Individual Level

Organizational Level

Volume

X

X

Weighted Volume

X

X

Intensity

X

X

Weighted Intensity

X

X

Frequency

X

X

Duration

X

X

Number of Episodes

X

X

X

X

Interaction Quantity

Interaction Intensity

Interaction Regularity

Mode of Interaction
Mode
Network Centrality
Number of Contacts
(Degree Centrality)

X

Normalized Degree
Centrality

X

Normalized Closeness
Centrality

X

of employee adaptiveness on service customization. Jayachandran et al. (2004) empirically estimated the customer knowledge
process on a six-item scale in order to determine its influence
on customer response capability. Finally, Joshi & Sharma (2004)
created a five-item scale to assess customer knowledge development and to evaluate its impact on new product performance.

5.2. Assessment of the Acquired Customer Intimacy

155

At the individual level, the following two assertions have been
derived from these scales to assess acquired customer knowledge:
1. My knowledge of [CustomerEmployeeName]’s needs is
thorough.
2. I learned a lot about [CustomerEmployeeName]’s preferences in the period I worked with him/her.
At the organizational level, the following three items have been
used:
1. My knowledge of [CompanyName]’s needs is thorough.
2. I learned a lot about [CompanyName]’s preferences in the
period I worked with it.
3. I know the customer [CompanyName] very well.
• Established Customer Relationships
The items used to empirically evaluate the relationships established with customers are inspired by the scales for assessing
relationship in a B2B context proposed by Crosby et al. (1990),
De Wulf et al. (2001), and Wuyts & Geyskens (2005). Crosby
et al. (1990) estimated relationship quality at the employee level
in order to determine its influence on services selling. De Wulf
et al. (2001) considered relationship quality as an outcome and
analyzed which factors have some influence on it such as interpersonal communication and preferential treatment. Last,
Wuyts & Geyskens (2005) investigated the formation of buyersupplier relationship and the influence of relationship quality
on partner selection.
Two items have been derived from these scales in order to assess the established customer relationships at the individual
level:
1. I have a high-quality relationship with [CustomerEmployeeName].
2. I have a very collaborative relationship with [CustomerEmployeeName].

156

5. CI Analytics Model and Methodology

At the organizational level, the following three items have been
used:
1. As an employee, I have a high-quality relationship with
[CompanyName].
2. As an employee, I have a very collaborative relationship
with [CompanyName].
3. I am satisfied with the relationship I have with [CompanyName].
The next section of this chapter elaborates on the metrics defined
to assess the leveraged customer intimacy components presented in
section 4.4.

5.3. Assessment of the Leveraged Customer
Intimacy
Section 4.4 elaborates on six leveraged customer intimacy components which reflect the actual benefits and competitive advantages
derived from the acquired customer intimacy. These six components
are: customization, customer loyalty, proactiveness, cross-selling, customer participation, and transaction costs reduction. Following the
approach of this thesis to assess customer intimacy upon customer
related data available in the provider’s information system, the six
parts of this section define multiple metrics in order to measure these
leveraged customer intimacy components.
While the metrics proposed by this thesis to assess the acquired customer intimacy cover both the individual and the organizational levels, the leveraged customer intimacy metrics solely focus on the organizational level for the following reasons. First, an assessment of
the leveraged customer intimacy at the individual level for each customer employee would not be useful in a B2B context as the value
proposition of the provider targets the customer organization rather
than the different customer employees. Second, the data which is
available for calculating the leveraged customer intimacy metrics

5.3. Assessment of the Leveraged Customer Intimacy

157

such as project records and sales results is specified at the organizational level only and not at the individual level. Thus, no data is
available to calculate the leveraged customer intimacy metrics at the
individual level. Table 5.2 presents the eight metrics proposed by
this thesis to assess the leveraged customer intimacy components.
Table 5.2.: Customer Intimacy Metrics for the Leveraged Customer
Intimacy
Customization
Customization Revenue Ratio
Proactiveness
Proactiveness Ratio

Loyalty
Customer Purchase Frequency
Ratio
Cross-Selling
Cross-Selling Revenue Ratio
Cross-Selling Diversity Ratio

Customer Participation
Customer Participation
Quantity

Transaction Costs Reduction
Transaction Effectiveness Ratio

Customer Participation Ratio

5.3.1. Customization
As explained in section 4.4.1, mass customization and customerization are different from customization in the context of customer intimacy. Therefore, proposed approaches for measuring mass customization and customerization such as those from Tu et al. (2007) or
Kumar (2005) are not suited in the scope of this thesis. In line with
the service customization through employee adaptiveness model, this thesis focuses on customization achieved by provider employees, such
as the completion of individual projects and the adaptation of existing solutions. It is possible to determine the revenues derived by
such projects and to compare them to those derived from standard
products and services such as software licenses in an IT context. This
thesis, thus, proposes to measure the degree of customization with

158

5. CI Analytics Model and Methodology

a customer as the ratio between the revenues generated from individually performed services for this customer and the total revenues
generated with this customer.
The overall duration T of the relationship between the provider and
the customer is divided in multiple time segments: T = {1, ..., i, .., n}.
Rstandard (i ) represents the revenues generated by standard products
and services in the time segment i and Rcustom (i ) the revenues generated by customized offering, such as project revenues in the same
time segment i. The customization revenue ratio can subsequently be
calculated as follows for the time segment i:
Customization Revenue Ratio(i) =

Rcustom (i )
Rstandard (i ) + Rcustom (i )

(5.36)

5.3.2. Customer Loyalty
Different approaches have been proposed in order to measure the
degree of loyalty of a customer. Dick & Basu (1994) propose to assess customer loyalty by means of a two-dimensional matrix. The
first dimension of this matrix – “repeat patronage” – indicates the
intention of the customer to repurchase products or services from
the same provider or brand. The second dimension of the matrix
– “relative attitude” – refers to the behavior of the customer with
regard to the provider organization and to the products and services he purchased from this provider. Dick & Basu (1994) argue
that loyalty is high when repeat patronage is high and when the relative attitude towards the provider is strongly favorable. Building
upon this research, Bennett & Bove (2001) propose to assess the recommendations and referrals performed by the customer to other
organizations as well as to consider the customer’s repurchasing behavior. In a similar way, Reichheld (2003, p.5) developed the metric
“net promoter score” and argues that this one dimensional construct
is strongly correlated with high loyalty. In order to assess the net promoter score, customers are asked to answer the following question
on a scale from 1 to 10: “How likely is it that you would recommend
our company to a friend or colleague?” Customers answering with a

5.3. Assessment of the Leveraged Customer Intimacy

159

value comprised between 9 and 10 are those which are very loyal to
the provider.
Consequently, this thesis proposes to apply these previously defined
approaches in order to assess the degree of customer loyalty. Three
different means are considered:
• Count the number of recommendations performed by customers
and more specifically those leading to additional revenues with
new customers. This information can be obtained if customers
inform the provider about the recommendations they performed,
or if prospective customers indicate that they contacted the provider on the recommendation of another company. Both solutions are easily implementable and can be fostered by a recommendation reward mechanism.
• Apply the net promoter score approach and survey customers
with regards to their intention to recommend the provider.
• Assess the behavior of the customer with regard to the frequency of his purchases over the past years. This information
can easily be derived from the sales results available in project
databases and in the CRM system. This solution is favored as it
corresponds to the analytical approach followed by this thesis.
More formally, the overall duration of the relationship between
the provider and the customer is defined as T and consists of
multiple time segments: T = {1, ..., i, ..., n}. The subset of time
segments in which the customer purchased some products or
services from the provider is denoted as U. Using the cardinalities of U and T, the customer purchase frequency ratio can be
calculated as follows:
Customer Purchase Frequency Ratio =

|U |
|T |

(5.37)

5.3.3. Proactiveness
In order to assess proactiveness, various empirical measurements
have been developed. Wallenburg et al. (2010) propose a four-item

160

5. CI Analytics Model and Methodology

scale for measuring proactive improvement behavior of service line
employees as perceived by customers in the context of B2B services.
This scale helps determining whether provider employees continuously perform process optimization, make suggestions for improvements, adapt the solution to the situation, and take initiatives. Frese
et al. (1997) associate proactiveness with the degree of personal initiative of provider employees and propose a scale to empirically assess
this degree.
Inspired by these approaches, this thesis propose an analytical means
to measure the degree of proactiveness by calculating the ratio of
proactive improvements to the total number of improvements performed over a specific time period. This information can easily be
retrieved from support and service management systems. Support
systems indicate the number of actions performed upon problems
identified by customers. Service management systems in addition
collect the number of change requests performed for each provided
solution. More formally, the overall duration of the relationship between the provider and the customer is defined as T and consists of
multiple time segments: T = {1, ..., i, ..., n}. If Pi and Ri represent
the number of improvements and changes performed to the solution
provided to the customer within the time segment i respectively at
the initiative of the provider and of the customer, the proactiveness
ratio for the time segment i can be calculated as follows:
Proactiveness Ratio(i) =

Pi
Pi + Ri

(5.38)

5.3.4. Cross-selling
Cross-selling has already been recognized as a key performance indicator in various industries such as in the finance sector (Kamakura
et al., 1991), and different approaches have been proposed in order
to assess cross-selling achievements from the perspective of the provider. Nash & Sterna-Karwat (1996) propose a methodology to assess cross-selling efficiency based on financial accounts details. Bauer
(2004, p.3) elaborates a customer cross-sell index in its KPI profiler

5.3. Assessment of the Leveraged Customer Intimacy

161

and suggests to calculate this index by “dividing the number of products sold by the number of customers purchasing a product in the
last two years.” However, these two approaches are not suited in the
context of this thesis as they do not allow an individual assessment
of the cross-selling performance for each customer. According to
Malms & Schmitz (2011, p.258), an effective measure of cross-selling
which considers customers on an individual basis has not yet been
proposed in past literature: “no prior studies conceptualize or operationalize cross-selling success in a way that accounts for the reliability
and validity of the measures.” They therefore suggest to measure the
effectiveness of the cross-selling activities as the “degree to which the
firm exploits customer’s full cross-selling potential.” However, this
assessment is not performed analytically, but empirically using a four
item Likert-type scale.

Consequently, inspired by the approach of Malms & Schmitz (2011),
this thesis proposes to create a revenue based metric in order to determine the cross-selling performance. Since cross-selling refers to
complementing the original offering to the customer with new products and services, this metric is based on the ratio between the revenues generated in a certain time segment by products and services that were already sold to the customer in the past and revenues
generated in the same time segment by products and services that
the customer purchases for the first time. The overall duration T of
the relationship between the provider and the customer is divided
in multiple time segments: T = {1, ..., i, .., n}. Pi represents the set
of products and services which are sold to the customer for the first
time in the time segment i. Qi represents the set of products and
services which (1) are sold to the customer within i and (2) were sold
to the customer before the beginning of i. If R( Pi ) and R( Qi ) represent the revenues generated by Pi and Qi over the time segment i, the
cross-selling revenue ratio for the time segment i can be calculated as
follows:
Cross-Selling Revenue Ratio(i) =

R( Pi )
R( Pi ) + R( Qi )

(5.39)

162

5. CI Analytics Model and Methodology

Another cross-selling metric based on the cardinalities of Pi and Qi
can also be calculated. This metric, which is defined as cross-selling
diversity ratio is more qualitative as it only considers the number of
products and services contained in Pi and Qi and it ignores the corresponding revenues:
Cross-Selling Diversity Ratio(i) =

| Pi |
| Pi | + | Qi |

(5.40)

5.3.5. Customer Participation
In order to assess the degree of customer participation, several empirical approaches have been proposed in past literature. Bettencourt
(1997) proposes to measure the degree of customer participation with
a four item Likert-type scale focusing on the willingness of the customer to share suggestions for improvement and problems. Cermak
& File (1994) used a one dimensional construct, asking the actual
level of involvement such as invested time and effort to determine
customer participation. Inspired by these approaches, this thesis suggests to use the number of proposed improvements by the customer
in a given time period in order to assess customer participation. This
information can be easily retrieved from support system in which
customer issues and requests are stored. More formally, considering the time segment i, and defining as Pi the set of improvements
proposed by the customer during i, the metric customer participation
quantity can be calculated as the cardinality of Pi :
Customer Participation Quantity(i) = | Pi |

(5.41)

This metric can be extended with a normalized version called customer participation ratio which considers the ratio between the improvements proposed by the customer and the revenues R(i ) generated with this customer in the time segment i:
Customer Participation Ratio(i) =

| Pi |
R (i )

(5.42)

5.3. Assessment of the Leveraged Customer Intimacy

163

5.3.6. Transaction Costs Reduction
A thorough literature review on existing approaches for measuring
transaction costs is proposed by Den Butter (2010, p.15), who argues
that “a considerable amount of research must be done” to quantify
the transaction costs, because most of the existing research has been
theoretical or qualitative. The main approach to assessing transaction costs outlined in Den Butter (2010) is to split the total costs of
the provider and the customer in production and transaction costs.
In line with this approach, but considering solely the provider’s perspective and limiting the scope of the transaction costs analysis to
the provider’s costs of sales, this thesis proposes to measure the
sales related time and resource investments performed by the provider for the customer and to put this value in relationship with the
corresponding revenues generated with this customer in the same
time period. According to Reichheld & Teal (2001) an organization
should invest less time and effort with customers it has established
relationships with to generate a certain transaction volume.
The time investments performed by provider employees to identify
sales opportunity and transform them into contracts can easily be
quantified if the customer-facing activities of the provider employees are tracked. For instance, if the interactions occurring with
customer employees such as meetings, phone calls, and letters are
stored in the CRM system, the overall customer interaction duration
can be measured and corresponds to the previously defined metric volume. This assessment can be extended with the sales related
but non-customer facing activities performed by the provider employees such as the preparation of customer meetings, the completion
of proofs of concept, and the answering of customer’s technical documents. More formally, if Ii represents the total interaction time
between sales employees of the provider and customer employees
within the time segment i, Ai the total amount of time spent by sales
employees of non-customer facing customer-related activity during
during the time segment i, and R(i ) the revenues generated with the
customer during the time segment i, the transaction effectiveness ratio

164

5. CI Analytics Model and Methodology

can be calculated as follows:
Transaction Effectiveness Ratio(i) =

Ii + Ai
R (i )

(5.43)

In summary, chapter 5 elaborated on the CI Analytics model and
methodology proposed by this thesis to assess customer intimacy.
A set of metrics has been conceived in order to assess each of the
customer intimacy components proposed in chapter 4 upon available data in the information system of the provider. Eight interaction
based metrics have been proposed to assess the acquired knowledge
of, and established relationships, with customers at the individual
and organizational levels. These metrics are inspired by past literature associating knowledge and relationships to the four interaction
characteristics quantity, intensity, regularity, and mode. In addition
to these eight metrics, three additional centrality metrics based on
the topology of the social network formed by provider and customer
employees have been used in order to assess the acquired customer
intimacy at the organizational level. Considering the competitive advantages and benefits derived from the customer intimacy strategy,
eight metrics based on interaction, activity, and revenue records have
been created upon existing literature in order to assess the values of
the six leveraged customer intimacy components identified in chapter 4. Furthermore, the CI Analytics methodology proposed in section 5.1 allows a calibration of the proposed customer intimacy metrics to the specific interaction patterns of the provider. This methodology is based on the established knowledge discovery in database
process (Fayyad et al., 1996a). The next chapter will detail the software implemented in the scope of this thesis to actually calculate
these metrics upon real data and to visualize them by means of a
graphical user interface.

Part III.

Evaluation

6. CI Analytics Software
The software CI Analytics has been conceived and implemented in
order to validate the CI Analytics model and methodology proposed
by this thesis in chapter 5. This software enables the calculation of
the customer intimacy metrics and makes them available to users by
means of a graphical user interface.
This application has been developed in cooperation with the IT software and services provider CAS Software AG, who markets the
customer relationship management (CRM) application CAS genesisWorld.1 More precisely, the software CI Analytics accesses the data
stored in CAS genesisWorld, such as interaction, project, and revenue data in order to calculate the customer intimacy metrics. Moreover, three students participated in the implementation of the software CI Analytics under the supervision of the author of this thesis:
Thomas Herzig worked on the elaboration of a first prototype for
the calculation of the acquired customer intimacy metrics such as
volume, intensity, and frequency of interaction. Johannes Kunze von
Bischhoffshausen complemented this prototype with the means to
calculate the leveraged customer intimacy metrics. Finally, Hakan
Bilgic participated in the analysis of the requirements from the end1

Further information on CAS genesisWorld is available at http://www.cas.de
(accessed on 29.09.2011).

168

6. CI Analytics Software

user perspective and in the development of the web-based user interface.
This chapter will elaborate on the technical details of the software
CI Analytics. Section 6.1 will analyze the business and technical requirements. Section 6.2 will set out the overall architecture of the
software CI Analytics and illustrate its user interface. Finally, section 6.3 will assess this application with regard to the previously
determined requirements and outline the results of a survey on the
potential business benefits of this application.

6.1. CI Analytics Business Analysis
In order to conceive and implement the software CI Analytics, a requirement analysis has been performed. This analysis is developed
in section 6.1.1. Subsequently, the relevant business objects for the
calculation of the customer intimacy metrics which are contained in
the database of the application CAS genesisWorld have been determined and are presented in section 6.1.2.

6.1.1. Requirements Analysis
This section elaborates on the requirements which have been considered for the implementation of the software CI Analytics. Following the approach of Sommerville (2007, p.119), this analysis distinguishes functional requirements reflecting the services and behaviors
that the system should provide from non-functional requirements
which are the constraints on the services provided by the system.
In addition, the requirements have been grouped in three distinct
domains, each of them covering a specific aspect of the application:
• Data Source Access
This domain covers requirements related to the access to data
containing elements of evidence that a certain degree of customer intimacy has been reached between a provider and a
customer. According to the CI Analytics model proposed in
chapter 5 (see figure 5.2), these data sources cover interaction

6.1. CI Analytics Business Analysis

169

and activity records, project information as well as financial details on the sales results.
• Customer Intimacy Calculation
This domain covers more specifically the requirements on the
calculation of the acquired and leveraged customer intimacy
metrics proposed in chapter 5.
• Customer Intimacy Representation
Requirements related to the customer intimacy representation
focus on end-users expectations with regard to the graphical
user interface (GUI) designed and implemented in the course of
this thesis to support the visualization of the customer intimacy
metrics.
This requirement analysis has been performed by investigating the
behavior of CAS employees with regard to their usage of the application CAS genesisWorld and of their potential usage of the software
CI Analytics. It considers the end-users perspective as well as the perspectives of the different IT functions that would be responsible for
administrating and supporting the software CI Analytics. This analysis has led to 15 functional and non-functional requirements which
are described in the next parts of this section. Table 6.1 provides a
summary of these requirements.
Table 6.1.: Functional and Non-Functional Requirements on CI Ana-

lytics

No.

Name

Requirement Type
Functional

Data Source Access
1

Access and process data stored in the
application CAS genesisWorld

X

2

Support the access to additional data
sources

X

NonFunctional

170

6. CI Analytics Software

Functional and Non-Functional Requirements on CI Analytics
(continued)

No.

Name

Requirement Type
Functional

NonFunctional

3

Minimize performance impact on CAS
genesisWorld

X

4

Provide scalable algorithm to access the data

X

5

Ensure that sensitive data is securely
handled

X

Customer Intimacy Calculation
6

Consider calibration parameters to perform
the metrics calculation

X

7

Calculate the acquired customer intimacy
metrics at the individual level

X

8

Calculate the acquired customer intimacy at
the organizational level, including the
network based centrality metrics

X

9

Calculate the leveraged customer intimacy
metrics

X

10

Use efficient algorithms and scalable
architecture to calculate the customer
intimacy metrics

X

11

Incrementally update the customer intimacy
metrics values

X

Customer Intimacy Representation
12

Visualize the acquired customer intimacy
metrics at the individual level by means of a
graph representation

X

13

Visualize the acquired customer intimacy
metrics at the organizational level

X

14

Visualize the leveraged customer intimacy
metrics by means of charts

X

6.1. CI Analytics Business Analysis

171

Functional and Non-Functional Requirements on CI Analytics
(continued)

No.

Name

Requirement Type
Functional

15

Provide data visualization by means of a
web-based interface

NonFunctional

X

6.1.1.1. Data Source Access
The following functional requirements related to the access to data
sources have been determined:
1. Access and process data stored in the application CAS genesisWorld
In order to validate the proposed approach of this thesis, the
customer intimacy metrics are calculated upon the data stored
in the application CAS genesisWorld. Thus, the software CI
Analytics should be able to access the underlying database of
CAS genesisWorld and process its data in a way that enables
the calculation of the customer intimacy metrics.
2. Support the access to additional data sources
Even though the software CI Analytics primarily focuses on data
contained in CAS genesisWorld, its architecture should provide
the ability to easily incorporate other sources of data such as
other CRM systems, project databases, groupware, or social
platforms. Thus, the architecture of the software CI Analytics
should be structured in a way that the data integration is separated from the actual calculation of the customer intimacy
metrics.
In addition, the following non-functional requirements have been determined:

172

6. CI Analytics Software

3. Minimize performance impact on CAS genesisWorld
CAS genesisWorld is a business critical application used by all
customer-facing employees such as sales representatives, services employees, and project managers. It is therefore mandatory that the software CI Analytics does not significantly impact
the performance of CAS genesisWorld: the access to the data
and the metrics calculation should be transparent to the endusers of CAS genesisWorld. Thus, the connections between
CI Analytics and CAS genesisWorld should be minimized and
efficiently performed. It should be possible to complete the
resource-intensive customer intimacy calculation on a separate
computer, and the resulting customer intimacy metrics should
be stored in their own database, externally to the CAS genesisWorld database.
4. Provide scalable algorithm to access the data
In order to calculate the customer intimacy metrics for a specific customer, the software CI Analytics must retrieve all interactions such as emails, meetings, phone calls, projects activities, and sales results related to this customer and stored in
CAS genesisWorld. Considering the potentially high number
of employees involved in the relationship between the provider
and customer, and the duration of this relationship which may
span over several years, these records can lead to multiple gigabytes of data. The algorithms for retrieving the data stored
in CAS genesisWorld must, therefore, be efficient and scalable
in order to handle large amounts of data.
5. Ensure that sensitive data is securely handled
CAS genesisWorld contains sensitive customer related information such as project information, sales results, and specific interaction records between provider and customer employees.
This information must be carefully managed to ensure that the
data is solely used for the purpose of calculating the customer
intimacy metrics. In addition the data access rights specified
in CAS genesisWorld should be propagated to the software CI
Analytics to make sure that the data is only accessed with the
appropriate credentials.

6.1. CI Analytics Business Analysis

173

6.1.1.2. Customer Intimacy Calculation
The following functional requirements related to the calculation of
the customer intimacy metrics have been identified:
6. Consider calibration parameters to perform the metrics calculation
Multiple parameters have been determined in chapter 5 in order to enable a calibration of the CI Analytics model to the specific patterns of each provider. These parameters are the time
period, the segment size, the interaction duration threshold,
the interaction quantity threshold, and the weighted interaction quantity threshold. The software CI Analytics should provide the ability to specify the values of these parameters as well
as to consider them in the calculation of the customer intimacy
metrics.
7. Calculate the acquired customer intimacy metrics at the individual level
In chapter 5, eight metrics have been conceived upon the concepts of customer interaction time and weighted customer interaction time in order to determine the acquired customer intimacy at the individual level. These metrics are for instance
volume, intensity, and frequency of interaction. Thus, the software CI Analytics should be able to (i) retrieve all provider
employees involved in the relationship with a specific customer,
(ii) retrieve all customer employees involved in this relationship, (iii) calculate the customer interaction time and weighted customer interaction time for each provider-customer employee combination, and (iv) calculate the eight corresponding
customer intimacy metrics for each of these combinations.
8. Calculate the acquired customer intimacy metrics at the organizational level, including the network-based centrality metrics
The software CI Analytics should be able to aggregate the customer interaction time and weighted customer interaction time
calculated at the individual level in order to determine the
values of the customer intimacy metrics at the organizational

174

6. CI Analytics Software

level. In addition, the software CI Analytics should include a
graph technology allowing the calculation of the centrality metrics degree centrality, normalized degree centrality, and normalized closeness centrality.
9. Calculate the leveraged customer intimacy metrics
Eight metrics have been proposed in chapter 5 to assess the
leveraged customer intimacy components such as the customization revenue ratio, the customer purchase frequency ratio, and
the cross-selling revenue ratio. The software CI Analytics should
be able to determine the values of these customer intimacy metrics for all customers of the provider for different time frames
such as the past quarter or the past year.
The non-functional requirements related to the calculation of the customer intimacy metrics are the following:
10. Use efficient algorithms and scalable architecture to calculate
the customer intimacy metrics
A high quantity of data, up to multiple thousands of interaction, project, and sales records has to be processed and evaluated in order to calculate the customer intimacy metrics at the
individual and organizational levels. Therefore, CI Analytics
must use efficient algorithms and rely on a scalable architecture in order to process this data and calculate the customer
intimacy metrics.
11. Incrementally update the customer intimacy metrics values
In order to take advantage of the proposed customer intimacy
metrics, this information must be precise and up-to-date. Therefore, the metrics should be automatically recalculated on a periodical basis, taking into account the most recent data such as
the last emails or the sales achievements stored in CAS genesisWorld. The calculation frequency impacts the number of access
to the data sources and, therefore, should be configurable. For
instance, the calculation of the customer intimacy metrics could
occur on a daily, weekly, or monthly basis.

6.1. CI Analytics Business Analysis

175

6.1.1.3. Customer Intimacy Representation
The following functional requirements have been established with regard to the representation and visualization of the customer intimacy
information:
12. Visualize the acquired customer intimacy metrics at the individual level by means of a graph representation
In order to graphically depict the values of the acquired customer intimacy metrics at the individual level, the software CI
Analytics should provide a graph based representation of all relationships established by provider employees with customer
employees, as illustrated in figure 1.1. In this graph, the nodes
should represent the provider and customer employees, and the
edges should reflect the customer intimacy established between
the corresponding employees. In addition, CI Analytics should
provide the ability to specify the customer intimacy metric used
to determine the weights of the edges on the graph as well as
the considered time period for the calculation of the metrics.
13. Visualize the acquired customer intimacy metrics at the organizational level
The software CI Analytics should provide the ability to visualize
the eight acquired customer intimacy metrics at the organizational level.
14. Visualize the leveraged customer intimacy metrics by means
of charts
The software CI Analytics should provide the ability to visualize
the eight leveraged customer intimacy metrics in the form of
column or line charts, thereby representing the evolution of the
metrics over time.
15. Provide data visualization by means of a web-based interface
In order to access the calculated customer intimacy metrics, a
web-based interface should be provided which allows a remote
access via Internet to the information. This interface should
include all information related to the acquired and leveraged

176

6. CI Analytics Software

customer intimacy and its development should adhere to technology standards. It should also provide the ability to select a
customer in a list and to represent the customer intimacy information for this specific customer only. It should in addition
provide the ability to filter and sort the displayed information.
Since the software CI Analytics is in its current state a prototypical implementation serving research purposes, the following nonfunctional requirements are out of the scope of this thesis: allowing a customization of the interface, implementing an authentication
mechanism to access the application, and guarantying specific service quality levels such as response time and availability.

6.1.2. Business Objects Analysis
Even though future versions of the software CI Analytics should provide the ability to retrieve data from various sources containing relevant information for the calculation of the customer intimacy metrics, the current version of CI Analytics focuses on the data contained
in the application CAS genesisWorld. This section outlines the data
retrieved by the software CI Analytics from the application CAS genesisWorld in order to calculate the customer intimacy metrics.
In CAS genesisWorld, the business objects of type Address, represent
either a customer organization, a customer employee, or a provider
employee. This is a central item in the architecture of CAS genesisWorld as it contains all customer related details such as names and
addresses. In addition to the business objects of type Address, the
analysis of the CAS genesisWorld database has led to the identification of nine relevant business objects in the context of this thesis.
These nine business objects are presented in table 6.2 and can be
categorized in three main categories:
• Interaction Business Objects record the interactions that occurred between the provider and customer employees. They
are required to calculate the customer interaction time and weighted customer interaction time which in turn are used to calculate the acquired customer intimacy metrics such as volume,
intensity, and frequency of interaction.

6.1. CI Analytics Business Analysis

177

• Activity Business Objects record the activities performed by
the provider employees. They are required as input to the leveraged customer intimacy metrics to assess the time spent by provider employees on customer projects and on the resolution of
customer problems.
• Revenue Business Objects track the details on the monetary
and non-monetary revenue generated with customers. The monetary revenue reflects the sales transaction achievements and
the non-monetary revenue concerns other form of value provided by the customer such as the customer’s suggestions for
improvements.
Table 6.2.: CI Analytics Business Objects
No.

Type

Description

Address Business Objects
1
Address
Addresses represent customer organizations,
customer employees, and provider employees in
the application CAS genesisWorld. Each address
record contains the required contact information
such as name, addresses, phone numbers, as
well as his preferences, his preferred contact, and
links to past activities.
Interaction Business Objects
2
Email
Exchanged emails with the customer can be automatically stored in CAS genesisWorld and, thus,
are available for the calculation of customer intimacy metrics. Details such as the sender and
receivers of the emails, time stamps, and content
can be retrieved.
3
Appointment
Appointments consist of the meetings organized
with the customer. They can be entered directly
in CAS genesisWorld or retrieved from calendar
applications. Details such as the list of participants to meetings as well as the date and duration of meetings can be retrieved.

178

6. CI Analytics Software

CI Analytics Business Objects (continued)
No.

Type

Description

4

Phone Call

5

Document

CAS genesisWorld can integrate a business
phone system, thereby allowing to store details
on the phone calls happening between provider
and customer employees, such as the provider
and customer employees names (through a mapping based on their phone numbers), the phone
calls dates and durations.
Documents refer to the letters received from, and
sent to, customer employees. Sender and receivers names as well as the document date are
available in each document record.

Activity Business Objects
6
Project Activity Customer projects are decomposed in multiple
project activities which can be accessed in order
to assess the tasks performed and time spent by
provider employee on customer projects. The activity type, list of involved provider employees,
date, and duration are available in each project
activity record.
7
Service Ticket
When a customer has a specific request or if
he experiences an issue with the provided solution, a service ticket is created and managed by
the support team until its closure. Service tickets are, thus, indicators of the efforts performed
by the provider to support the customer. The
names of the customer and involved customer
and provider employees as well as the date, the
time spent, and the actions undertaken to solve
the problem are available in each service ticket
record.
Revenue Business Objects
Invoice Line
8
Invoice Line Items provide details on the prodItem
ucts and services purchased by the customer,
such as the product and service references, the
price paid, and the date of purchase.

6.2. CI Analytics Architecture

179

CI Analytics Business Objects (continued)
No.

Type

Description

9

Suggestion

Suggestions are special types of service tickets in
which customers report suggestions for improvement. They are thus, to some extent, indicators of
the degree of participation of the customer to the
development and improvement of the solution.

6.2. CI Analytics Architecture
6.2.1. Architecture Overview
This section presents an overview of the architecture of the software
CI Analytics. This architecture relies on a data warehouse and on an
extract, transform, load (ETL) component for extracting and processing
the data stored in operational databases. It adheres, thus, to the
architectural standards for decision support applications proposed
by Turban et al. (2011). Figure 6.1 depicts the main components of
this architecture as well as the interface of the software CI Analytics
with an external customer data source such as CAS genesisWorld.
This architecture is structured along the three following layers:
• Data Layer (CI Data Warehouse)
This layer contains the CI Data Warehouse, which is the underlying database of the software CI Analytics, as well as the operational database of the considered source of customer intimacy
to calculate the customer intimacy metrics, in particular, the
database of the application CAS genesisWorld. CI Data Warehouse, as suggested by its name, is an implementation of a data
warehouse. Turban et al. (2011, p.328) define a data warehouse
as a “repository of current and historical data of potential interest to managers throughout the organization” and precise
that its data is structured in a way that supports analytical processing activities. CI Data Warehouse is not used for operational

180

6. CI Analytics Software

purposes, but solely stores customer related data which is relevant for the calculation of the customer intimacy metrics. It
is also optimized for efficiently reading and processing large
amounts of data. Further details on the CI Data Warehouse are
provided in section 6.2.2.
• Application Layer (CI ETL and CI Services)
This layer reflects the business logic components of the software
CI Analytics, namely the CI ETL and the CI Services:
– The component CI ETL populates the database CI Data
Warehouse based on the data available in operational databases such as the CAS genesisWorld database. CI ETL
implements an extract, transform, and load process (Turban
et al., 2011, p.344). It reads the data from the operational
database, filters the relevant records, transforms them to
prepare the calculation of the customer intimacy metrics,
and loads the transformed data in the CI Data Warehouse
database. This component is further developed in section 6.2.3
– The CI Services are the clients of the database CI Data Warehouse. They expose the functionality of calculating the
different customer intimacy metrics upon the data available in CI Data Warehouse by means of RESTful web services. RESTful web services use the standard http protocol
and adhere to the REST – Representational State Transfer
– architecture which simplifies components interoperability, increases scalability, and provides an easy access to
the customer intimacy metrics (Richardson & Ruby, 2007).
Section 6.2.4 elaborates on the CI Services.
• Presentation Layer (CI Dashboard)
This layer focuses on the presentation of the customer intimacy
information to the users and consists of the component CI Dashboard. The CI Dashboard is the graphical user interface (GUI)
implemented in the software CI Analytics. It provides a graphbased representation of the acquired customer intimacy and a
chart-based representation of the leveraged customer intimacy.

6.2. CI Analytics Architecture

181

Users can input the necessary parameters such as customer
name, time frame and requested metrics in order to configure
the CI Services calls. The CI Dashboard is a web-based graphical
user interface, letting users access the information with their
web browser from their organization’s intranet or from Internet. Further information on this component is proposed in section 6.2.5.
Operational System

CI Analytics

(CAS genesisWorld)

CI Dashboard
Enterprise
Information System

CI ETL

Operational Data

CI Services

CI Data Warehouse

Figure 6.1.: CI Analytics Architecture
The next sections detail each of the previously mentioned components of the software CI Analytics.

6.2.2. CI Data Warehouse
The CI Data Warehouse is the underlying database of the software
CI Analytics. It stores data in a form allowing the calculation of
the customer intimacy metrics after it has been extracted from the
different operational sources of customer intimacy data, such as the
operational database of the application CAS genesisWorld. A key
aspect of data warehouses is that they are subject-oriented and multidimensional (Turban et al., 2011, p.332). This means that the data
is organized along specific subjects and it is structured in a way that
supports its analysis along multiple dimensions. In order to fulfill
these characteristics, the design of data warehouse tables follows the

182

6. CI Analytics Software

star schema which consists of a central subject-focused fact table surrounded by multiple dimension tables (Turban et al., 2011, p.351).
Following this approach, the fact tables in the CI Data Warehouse focus on the key elements required to perform the calculation of the
customer intimacy metrics. Three fact tables have been conceived:
• Customer interaction time fact table
The customer interaction time fact table stores details on the
interaction time spent with each customer employee. Each interaction business object stored in CAS genesisWorld such as
emails, meetings, phone calls, and documents are transformed
by the CI ETL process into a record in the customer interaction
time fact table. This record contains the duration of the interaction act to allow a calculation of customer interaction time
and weighted customer interaction time as well as additional
dimensional information. In order to provide multiple dimensional analyses of the customer interaction time, the following
dimension tables have been conceived:
– CustomerCompany, to focus on a specific customer organization,
– CustomerEmployee, to focus on a specific customer employee,
– ProviderEmployee, to focus on a specific provider employee,
– Date, to focus on a specific time frame,
– Channel, to focus on a specific interaction channel,
– Project, to focus on a specific project.
Figure 6.2 illustrates the customer interaction time fact table
surrounded by its six dimension tables. Using these seven tables it is possible to combine multiple pieces of dimensional information to calculate the customer interaction time, weighted
customer interaction time, and subsequently the acquired customer intimacy metrics for very specific criteria. For instance,
it is possible to calculate the acquired customer intimacy metrics for the provider employee p with the customer employee

6.2. CI Analytics Architecture

183

c from the customer organization C within a specific year. This
analysis could be further detailed by specifying a channel of
interaction or a project reference.
ProviderEmployee
PK

EmployeeID
Name
ChristianName

Date
PK

Date

CustomerEmployee
PK

DateID

EmployeeID
FactTable
Name
ChristianName

PK
PK
PK
PK
PK
PK
PK

CustomerCompanyID
EmployeeID
DateID
ProjectID
ChannelID
CustomerEmployeeID
ProviderEmployeeID
CIT

PK

ChannelID
Channelname

CustomerCompany
PK

Channel

CustomerCompanyID
Companyname
Project
PK

ProjectID
Projectname

Figure 6.2.: Customer Interaction Time Star Schema
• Customer activity time fact table
The customer activity time fact table focuses on the duration of
the activities performed by provider employees for customers.
Similarly to the interaction business objects, the business objects of type project activity or service ticket are transformed by
the CI ETL process into records in the customer activity time
fact table. Records in this table contain the activity duration
as well as dimensional values corresponding to the six dimensional tables surrounding the customer activity time fact table.
Five of these dimension tables are the same as those of the customer interaction time fact table, namely: CustomerCompany,

184

6. CI Analytics Software

CustomerEmployee, ProviderEmployee, Date, and Project. However, the dimension table ActivityType replaces the dimension
table Channel. The dimension ActivityType allows to evaluate
whether the activity is value-adding like a consulting task, or
non-value-adding such as an administrative task.
• Customer value return fact table
Records in the customer value return fact table contain details
on the monetary and non-monetary revenues generated with
the different customers of the provider. Business objects of
type invoice line item represent the monetary revenues generated with customers such as the achieved sales transactions.
They are, thus, converted into records containing their monetary value in the customer value return fact table. Business objects of type suggestion represent a special form of non-monetary customer value return as the information provided in the
suggestion can be used by the provider to improve its value
proposition, thereby enabling him to achieve a new competitive
advantage. Therefore, the business objects of type suggestion
are also transformed into records in the customer value return
fact table. In the current version of the software CI Analytics, the
business objects of type suggestion are converted into records
having a constant monetary value. Its architecture, however,
would support a monetary quantification of the customer suggestions which could be elaborated in future research. The dimensional information provided in the records of the customer
value return fact table allows to distinguish monetary revenues
derived from invoice line items from the non-monetary revenues which are derived from suggestions. It also allows the
calculation of the leveraged customer intimacy metrics for specific customers, time periods, or projects. Four dimension tables therefore surround the customer value return fact table:
– CustomerCompany, to focus on the revenues generated with
a specific customer organization,
– Project, to focus on the revenues derived from a specific
project,

6.2. CI Analytics Architecture

185

– Date, to estimate the revenues in a specific time frame,
– ValueSource to determine whether the record refers to monetary or non-monetary revenue.
In order to implement the CI Data Warehouse, the application Microsoft SQL Server 2008 R2 has been used.2 This application was chosen because it provides the required tools to realize a data warehouse
upon standard database management functions, thereby decreasing
the complexity of the overall software architecture, and because this
is the default database of the application CAS genesisWorld.

6.2.3. CI ETL
The component CI ETL implements the extract, transform, and load
process of the software CI Analytics. It is responsible for populating
the database CI Data Warehouse. In the extraction phase, data which
is relevant for the calculation of the customer intimacy metrics is
read out of the operational databases, such as the database of CAS
genesisWorld. During the transformation phase, this data is filtered
and converted into the format of the CI Data Warehouse in order to
be entered in one of the fact tables or one of the dimension tables.
Finally, during the loading phase, the transformed data is actually
stored in the fact and dimension tables of the CI Data Warehouse.
The process of the CI ETL component consists of eight subprocesses
which are depicted in figure 6.3:
1. ETL CustomerCompany Data extracts customer organizations details stored in the business objects of type address such as
the names and reference numbers of the companies. It subsequently transforms and loads them in the dimension table
CompanyName.
2. ETL CustomerEmployee Data extracts data related to customer
employees which is stored in the business objects of type address. Then, it transforms and loads it in the dimension table
CustomerEmployee.
2

Further details are available at
http://www.microsoft.com/sqlserver/en/us/default.aspx (accessed on 29.09.2011).

186

6. CI Analytics Software

3. ETL ProviderEmployee Data extracts details on provider employees, which are also stored in the form of business objects of
type address. These objects are subsequently transformed and
loaded in the dimension table ProviderEmployee.
4. ETL Project Data extracts details on customer projects which are
stored in the business objects of type project activity. It then
transforms and loads them in the dimension table Project.
5. ETL Activity Data extracts details on the activity durations from
the business objects of type project activity or service ticket.
This information is then used to populate the customer activity
time fact table .
6. ETL Revenue Data extracts financial information out of the business objects of type invoice line item. Subsequently, it transforms this data and loads it as facts into the customer value
return fact table.
7. ETL Customer Participation Data filters the business objects of
type service request which are specifically referring to customer
suggestions, then transforms and loads them as facts in the
customer value return fact table. In the current version, as explained in section 6.2.2, these specific facts all have the same
monetary value and can be differentiated from the monetary
customer revenues derived from business objects of type invoice line items through the ValueSource dimension.
8. ETL Interaction Data extracts the interaction duration as well
as the required dimensional information from the interaction
business objects of type email, appointment, phone call, or document. Then, it transforms this data and loads it into the customer interaction time fact table.
The implementation of the component CI ETL has been performed
with the Microsoft SQL Server Integration Services since both the CI
Data Warehouse and the database of CAS genesisWorld are realized
with Microsoft SQL Server 2008 R2.3
3

Further details are available at
http://www.microsoft.com/sqlserver/en/us/solutions-technologies/business-

6.2. CI Analytics Architecture

187

CAS genesisWorld
Business Objects

CI Data Warehouse
Dimension Tables

Address

1

CustomerCompany

Project Activity

2

CustomerEmployee

Service Ticket /
Suggestion

3
ProviderEmployee

4
Invoice Line Item

Project

Email

5

Appointment

1.
2.
3.
4.

6

Phone Call

7

Document

8

ETL Customer Company Data
ETL Customer Employee Data
ETL Provider Employee Data
ETL Project Data

5.
6.
7.
8.

Fact Tables
Customer Activity
Time
Customer Revenue
Value
Customer
Interaction Time

ETL Activity Data
ETL Revenue Data
ETL Participation Data
ETL Interaction Data

Figure 6.3.: Overview of the CI ETL Process

6.2.4. CI Services
The CI Services provide client applications such as the CI Dashboard
with an access to the data stored in the CI Data Warehouse and expose the functionality of calculating the customer intimacy metrics
by means of standardized RESTful web services. They thereby enable the calculation of the acquired and leveraged customer intimacy
metrics proposed in chapter 5.
Following the RESTful web services approach, client applications invoke the CI Services with a GET request containing the required input
parameters such as the name of the metric and the considered time
frame to calculate the customer intimacy metrics. The CI Services
intelligence/integration-services.aspx (accessed on
29.09.2011).

188

6. CI Analytics Software

subsequently convert these requests into a set of SQL queries, perform these queries on the data contained in the CI Data Warehouse,
aggregate the results, perform the calculation of the customer intimacy metrics, and finally return the results to the client application
in an XML format. Technical details such as the input and output
parameters of the CI Services are presented in appendix E.
Each CI Service in the software CI Analytics is designed to calculate
one of the acquired and leveraged customer intimacy metrics proposed in chapter 5. Thus, the following CI Services have been conceived and implemented:
• CI Services for the Acquired Customer Intimacy Metrics
In section 5.2, this thesis establishes eight metrics to assess the
acquired customer intimacy upon the concept of customer interaction time and weighted customer interaction time, namely
volume, weighted volume, intensity, weighted intensity, frequency, duration, number of episodes, and mode of interaction.
Moreover, two levels of analysis have been proposed:
– The individual level of analysis allows an assessment of
the degree of customer intimacy established between provider and customer employees. Consequently, eight services have been conceived to assess the acquired customer
intimacy metrics at the individual level. These services
take as input the reference to a customer organization, the
beginning and end dates of the chosen calculation time period, and the calibration parameter values specified in appendix E.1. They return a graph in the DyNetML format
proposed by Tsvetovat et al. (2004). This graph represents
the social network formed by the provider and customer
employees: its nodes symbolize the employees, and the
weights on the edges of the graph are the actual customer
intimacy metrics values.
– The organizational level of analysis considers the degree
of customer intimacy established between a provider employee and a customer organization. Consequently, eight

6.2. CI Analytics Architecture

189

services have been realized in order to implement the calculation of the eight acquired customer intimacy metrics
at the organizational level. These services take as inputs
a reference to a customer organization, the beginning and
end dates of the chosen calculation time period, different
calibration parameters which are specified in appendix E.1
as well as the reference to the provider employee for which
the metric is calculated. These CI Services return the values of the considered customer intimacy metrics. These
eight services are presented in table 6.3 and further detailed in appendix E.1. At the organizational level of analysis, in addition to the eight customer interaction time
based metrics, three network centrality metrics have been
conceived, namely the number of contacts (degree centrality), the normalized degree centrality, and the normalized
closeness centrality. These services have not yet been implemented in the software CI Analytics. However, they
have been implemented in the first CI Analytics prototype
called CI Graph. Appendix E.3 provides additional details
on CI Graph.
• CI Services for the Leveraged Customer Intimacy Metrics
Eight customer intimacy metrics have been proposed in section 5.3 in order to assess the leveraged customer intimacy
components. These metrics are: customization revenue ratio,
customer purchase frequency ratio, proactiveness ratio, crossselling revenue ratio, cross-selling diversity ratio, customer participation quantity, customer participation ratio, and transaction effectiveness ratio. With the exception of the metric proactiveness ratio for which no data is available in the application CAS genesisWorld, all leveraged customer intimacy metrics have been implemented in a CI Service. Thus, seven services have been realized in the software CI Analytics. Table 6.3
presents these seven services and the corresponding customer
intimacy metrics. Further details such as the inputs and outputs of the services are provided in appendix E.2.

190

6. CI Analytics Software

The Windows Communication Foundation4 which is part of the Microsoft .NET framework has been used in order to implement the
CI Services as resource-oriented REST services (Chappell, 2010). This
technology has been chosen because it provides an easy integration
with the underlying database of the CI Date Warehouse Microsoft SQL
Server 2008 R2. Since the created services are available via the standard http protocol, these services are not constrained into the .NET
environment but can be accessed by any application supporting the
http protocol. The actual development has been performed with the
software Microsoft Visual Studio in the programming language C#.5
Table 6.3.: CI Services Overview
CI Service

Customer Intimacy Metric
Individual Level

Organizational Level

Volume

Volume Service

Org Volume Service

Weighted Volume

WVolume Service

Org WVolume Service

Intensity

Intensity Service

Org Intensity Service

Weighted Intensity

WIntensity Service

Org WIntensity Service

Frequency

Frequency Service

Org Frequency Service

Duration

Duration Service

Org Duration Service

Number of Episodes

NumberEpisodes
Service

Org NumberEpisodes
Service

Mode

Mode Service

Org Mode Service

Acquired Customer
Intimacy

4
5

Number of Contacts (Degree
Centrality)

Available in first
prototype only

Normalized Degree
Centrality

Available in first
prototype only

Further details are available at
http://msdn.microsoft.com/en-us/netframework/aa663324 (accessed on 29.09.2011).
Further details are available at
http://msdn.microsoft.com/en-us/vcsharp/aa336809 (accessed on 29.09.2011).

6.2. CI Analytics Architecture

191

Table 6.3.: CI Services Overview (Continued)
CI Service

Customer Intimacy Metric
Individual Level
Normalized Closeness
Centrality

Organizational Level
Available in first
prototype only

Leveraged Customer
Intimacy
Customization Revenue
Ratio

Customization Revenue
Ratio Service

Customer Purchase
Frequency Ratio

Customer Purchase
Frequency Ratio
Service

Proactiveness Ratio
Cross Selling Revenue Ratio

CrossSelling Revenue
Ratio Service

Cross-Selling Diversity
Ratio

CrossSelling Diversity
Ratio Service

Customer Participation
Quantity

Customer Participation
Quantity Service

Customer Participation
Ratio

Customization
Participation Ratio
Service

Transaction Effectiveness
Ratio

Transaction
Effectiveness Ratio
Service

6.2.5. CI Dashboard
The CI Dashboard provides the means to graphically visualize the
acquired and leveraged customer intimacy metrics. In order to facilitate its adoption, the CI Dashboard has been implemented as a webapplication which is accessible with an Internet browser. Figure 6.4
illustrates the main interface of the CI Dashboard with fictive data.

192

6. CI Analytics Software

Figure 6.4.: Main Interface of the CI Dashboard

6.2. CI Analytics Architecture

193

The bottom part of the interface allows users to specify the name
of a customer, the first and last years of the considered time period,
and the metric to be displayed on the edges of the social network
representation (volume in figure 6.4). After clicking on the Update
button located in the bottom right corner, web services are called
in order to render the two parts of the CI Dashboard reflecting the
assessment of the acquired and leveraged customer intimacy:
• Acquired Customer Intimacy Visualization
The diagram on the left-hand part of the CI Dashboard provides
a representation of the social network formed by the provider
and customer employees. The rectangles symbolize the employees and are aligned in two rows: the rectangles in the top
row represent the provider employee and those in the bottom
row represent the customer employees. The edges on the diagram connect the provider employees to the customer employees. An edge between the provider employee p and the customer employee c indicates that the chosen acquired customer
intimacy metric calculated for the couple of employees { p, c}
on the specified time frame has a value greater than 0. The
weights of the edges which are displayed on the diagram indicate the actual values of the selected customer intimacy metric.
For instance, the metric volume has a value of 24.5 for the couple of employees {“Catherine Jones” ; “Sarah Lundberg”}. This
means that during the specified time period, the provider employee “Catherine Jones” interacted for a duration of 24.5 hours
with the customer employee “Sarah Lundberg”.
The CI Dashboard provides the functionality to zoom into the diagram by selecting an employee and using the slider on the top
left corner. Moreover when an employee is selected, a new rectangle is displayed, as illustrated in figure 6.5(a), in which detailed information on the provider employee can be displayed
such as his role and organization units. In the future, the CI
Dashboard should provide the ability to display the acquired
customer intimacy metrics at the organizational level in this
rectangle. Figure 6.5(b) illustrates the capability of the software

194

6. CI Analytics Software

CI Analytics and especially of the CI Dashboard to handle large
amount of data, thereby demonstrating the scalability of the
implemented algorithms. This diagram is based upon real data
which has been anonymized. This picture also emphasizes the
needs for additional filtering capabilities which should be provided in a future version.
• Leveraged Customer Intimacy Visualization
The right-hand side of the CI Dashboard illustrated in figure 6.4
provides the functionality to visualize the leveraged customer
intimacy metrics by means of chart diagrams. The current version of CI Dashboard allows a representation of three out of the
eight proposed customer intimacy metrics, namely customization revenue ratio, cross-selling revenue ratio, and customer
participation ratio. The CI Dashboard will be extended in the
next release to include the remaining leveraged customer intimacy metrics such as the customer purchase frequency ratio or
the transaction effectiveness ratio.
While the default representation of the customer intimacy metrics uses pie charts, column charts can also be displayed, as illustrated in figure 6.6 for the metric cross-selling revenue ratio.
Using this representation, it is possible to analyze the evolution
of customer intimacy metric value over time.
The CI Dashboard has been implemented with the technology Silverlight.6 Silverlight is an application framework enabling the creation and delivery of rich internet applications which can be installed
as a plug-in in the Internet Browser. This technology has been chosen because it provides a good integration with the Windows Communication Foundation used to implement the CI Services and it is a
technology used in the latest release of CAS genesisWorld.
In order to realize the graph representation supporting the visualization of the acquired customer intimacy metrics at the individual
level, the technology NodeConnect developed by Hodnick (2009) has
6

Further details are available at http://www.microsoft.com/silverlight (accessed on
24.09.2011).

6.2. CI Analytics Architecture

(a) Detailed Information

(b) Large Social Network

Figure 6.5.: CI Dashboard: Acquired Customer Intimacy

195

196

6. CI Analytics Software

Figure 6.6.: CI Dashboard: Leveraged Customer Intimacy
been chosen as it consists of a simple and free of charge library allowing an easy customization of the graph representation. In order to
realize the chart-based representation of the leveraged customer intimacy metrics, the Quickchart library developed by amCharts7 has
been chosen as this technology allows to fulfill the requirements established in section 6.1 while remaining free of charge and open source.
The next section of this chapter develops an evaluation of the software CI Analytics with regard to the previously defined requirements
and to the actual benefits for its users.

6.3. CI Analytics Evaluation
This section presents an evaluation of the software CI Analytics. Part
6.3.1 contains an analysis of the software CI Analytics with regard
7

Further details are available at http://wpf.amcharts.com/quick/ (accessed on
24.09.2011).

6.3. CI Analytics Evaluation

197

to the functional and non-functional requirements identified in section 6.1. Subsequently, part 6.3.2 introduces the results of an empirical survey performed to assess the businesss benefits of the software
CI Analytics.

6.3.1. Requirements Assessment
In section 6.1, 15 functional and non-functional requirements that
should be fulfilled by the software CI Analytics have been defined.
Table 6.4 summarizes to which extent these requirements have been
completed in the actual version of the software CI Analytics. As outlined in this table, all requirements have been at least partly achieved
and nine out of the 15 requirements are fully achieved. The next
paragraphs provide further details on each of these achievements:
Table 6.4.: Fulfillment

of
Requirements

the

Functional

and

Non-Functional

Data Source Access
1

Access and process data stored in the
application CAS genesisWorld (functional)

X

2

Support the access to additional data sources
(functional)

X

3

Minimize performance impact on CAS
genesisWorld (non-functional)

X

4

Provide scalable algorithm to access the data
(non-functional)

X

5

Ensure that sensitive data is securely handled
(non-functional)

X

Not achieved

Name

Partly Achieved

No.

Fully Achieved

Achievement

198

6. CI Analytics Software

Fulfillment of the Functional and Non-Functional Requirements
(Continued)

Customer Intimacy Calculation
6

Consider calibration parameters to perform
the metrics calculation (functional)

X

7

Calculate the acquired customer intimacy
metrics at the individual level (functional)

X

8

Calculate the acquired customer intimacy at
the organizational level, including the
network based centrality metrics (functional)

X

9

Calculate the leveraged customer intimacy
metrics (functional)

X

10

Use efficient algorithms and scalable
architecture to calculate the customer
intimacy metrics (non-functional)

11

Incrementally update the customer intimacy
metrics values (non-functional)

X
X

Customer Intimacy Representation
12

Visualize the acquired customer intimacy
metrics at the individual level by means of a
graph representation (functional)

13

Visualize the acquired customer intimacy
metrics at the organizational level (functional)

X

14

Visualize the leveraged customer intimacy
metrics by means of charts (functional)

X

X

Not achieved

Name

Partly Achieved

No.

Fully Achieved

Achievement

6.3. CI Analytics Evaluation

199

Fulfillment of the Functional and Non-Functional Requirements
(Continued)

15

Provide data visualization by means of a
web-based interface (functional)

X

Not achieved

Name

Partly Achieved

No.

Fully Achieved

Achievement

1. Access and process data stored in the application CAS genesisWorld (Fully Achieved)
The component CI ETL provides the ability to access all required data stored in CAS genesisWorld to calculate the customer intimacy metrics.
2. Support the access to additional data sources (Fully Achieved)
The modular architecture of the software CI Analytics based on
the CI ETL and the CI Data Warehouse allows to easily add new
sources of data by updating the CI ETL component and, if required, by adding new tables in the CI Data Warehouse.
3. Minimize performance impact on CAS genesisWorld (Fully
Achieved)
Once the data contained in CAS genesisWorld has been retrieved by the CI ETL component and loaded in the CI Data
Warehouse, the software CI Analytics is completely independent
from CAS genesisWorld and, thus, does not impact its performance. The CI ETL component can be scheduled to run during
time period where CAS genesisWorld is not used, for instance
during the night.

200

6. CI Analytics Software

4. Provide scalable algorithm to access the data (Fully Achieved)
The access to data is based upon standard business intelligence
methods, using an ETL process and a data warehouse, and it
relies on established technology. It is therefore capable of efficiently handling large amount of data. In a test, 20 minutes
were necessary to process data related to 14 customer organizations stored in a real CAS genesisWorld database, whereas the
first prototype introduced in appendix E.3 required 20 hours to
complete the same operation.
5. Ensure that sensitive data is securely handled (Partly Achieved)
The data in the CI Data Warehouse is securely stored as this component is protected by the security mechanisms implemented
in Microsoft SQL Server 2008. However, in the current version,
the different components CI ETL, CI Data Warehouse, CI Services
and theCI Dashboard do not communicate using secured protocols. In addition the CI Dashboard does not implement an
authentication mechanism. This aspect is, however, not in the
scope of this thesis.
6. Consider calibration parameters to perform the metrics calculation (Fully Achieved)
The CI Services take the calibration parameters such as the time
period or the time segment size as inputs and use them to calculate the customer intimacy metrics. Further details are provided in appendix E.
7. Calculate the acquired customer intimacy metrics at the individual level (Fully Achieved)
As outlined in table 6.3, eight CI Services provide the functionality to calculate the acquired customer intimacy at the individual
level.
8. Calculate the acquired customer intimacy at the organizational
level, including the network based centrality metrics (Partly
Achieved)
Eight CI Services provide the ability to calculate the eight customer interaction time based acquired customer intimacy metrics at the organizational level. The network-based centrality

6.3. CI Analytics Evaluation

201

metrics are, however, not yet implemented in the current version of the software CI Analytics. To ensure the completeness of
this thesis, they are implemented in the first prototype CI Graph
presented in appendix E.3.
9. Calculate the leveraged customer intimacy metrics (Partly
Achieved)
Seven CI Services provide the ability to calculate seven out of the
eight leveraged customer intimacy metrics. The metric proactiveness ratio is not implemented in the current version of CI
Analytics because CAS genesisWorld does not contain the relevant data for its calculation.
10. Use efficient algorithms and scalable architecture to calculate
the customer intimacy metrics (Fully Achieved)
The algorithms implemented in the CI Services are capable of
handling large amount of data and efficiently process the metrics calculation. Further details on these algorithms are available upon request from the author.
11. Incrementally update the customer intimacy metrics values
(Partly Achieved)
The CI ETL process can be scheduled to run automatically in
different time intervals. The configuration of the scheduler is,
however, not yet implemented in the CI Analytics. This could
be achieved in a future version via an additional administration
interface in the CI Dashboard.
12. Visualize the acquired customer intimacy metrics at the individual level by means of a graph representation(Fully Achieved)
As illustrated in figure 6.4, the CI Dashboard provides the functionality to represent the acquired customer intimacy metrics
in the form of a graph representation.
13. Visualize the acquired customer intimacy metrics at the organizational level (Partly Achieved)
Figure 6.5(a) shows that by selecting a specific employee on the
graph displayed by the CI Dashboard, a window containing additional employee related information is being displayed. This

202

6. CI Analytics Software

windows can be used to represent the acquired customer intimacy metrics at the organizational level. This feature is not
yet implemented in the current version of CI Analytics but it
is, however, available in the first prototype described in appendix E.3.
14. Visualize the leveraged customer intimacy metrics by means
of charts (Partly Achieved)
The current version of the CI Dashboard provides the graphical representation of three out of the eight leveraged customer
intimacy metrics. Its next version will include a graphical representation of the five remaining metrics.
15. Provide data visualization by means of a web-based interface
(Fully Achieved)
The CI Dashboard has been realized as a web-based interface
using the Silverlight technology. It is, therefore, remotely accessible with any Internet browser.

6.3.2. Business Benefits Evaluation
In order to assess the potential of the software CI Analytics, a survey
was conducted in collaboration with Thomas Herzig with 25 employees of three different IT software and services companies in July
2010. These participants were introduced to the project CI Analytics
and were shown screenshots of the first prototype of the software
CI Analytics which is described in appendix E.3. They were subsequently asked to evaluate the potential of this software with regard
to its potential business benefits.
The assessment was performed using the questionnaire presented in
appendix E.3. This questionnaire consists of 12 items which are assessed on five-point Likert-type scales.8 These items reflect the three
business benefits of the software CI Analytics outlined in section 1.3.
25 employees were surveyed and all filled in their questionnaires,
8

Further details on Likert-type scales are available in section 3.1.3

6.3. CI Analytics Evaluation

203

resulting in a 100% response rate. As detailed in appendix E.3 figure E.8, the participating employees have different roles and positions in their organization such as management, sales, services, or
development. They were all involved in customer facing activities
during the year preceding the survey: 88% of them were in contact
with more than three customer organizations, 72% of them were in
contact with more than 10 customer employees, and 64% spent over
20% of their time in customer related activities.
The three business benefits which have been considered in this survey are the following:
• Business Benefit 1: CI Analytics helps its users to gain an
overview of the relationships established with customers and
customer employees.
In order to evaluate the potential of this first benefit, the following two items were assessed:
– Question 5: I would use this overview to identify colleagues who have knowledge about the customer organization (strategy, process, organization, behavior, etc).
– Question 6: I would use this overview to identify colleagues who have established relationships with customer
employees.
The results to these two questions are presented in figures 6.7
and 6.8. They confirm the relevance of the model and methodology proposed by this thesis as over 90% of the participants
agree or strongly agree that they would use a software such as
CI Analytics to identify their colleagues who have some knowledge about customers, and 80% of them agree or strongly
agree that they would use it to identify colleagues who have
established relationships with customer employees.

204

6. CI Analytics Software

Figure 6.7.: CI Analytics: Business Benefit 1 – Question 5

Figure 6.8.: CI Analytics: Business Benefit 1 – Question 6
• Business Benefit 2: CI Analytics creates an awareness of the
business relationships established by provider employees and,
thus, fosters the exchange of customer related information
among provider employees.
To evaluate this second business benefit, the following items

6.3. CI Analytics Evaluation

205

were assessed by the participants:
– Question 7: This relationship network overview would
help us share knowledge about the customer inside our
organization.
– Question 8: This relationship network overview would
help us coordinate our activities towards the customer and
to be seen as one team by the customer.
Figures 6.9 and 6.10 outline the results of the assessment of
these two items. The results to question 7 confirm that the
graph representation of the social network formed by the provider and customer employees in CI Analytics is a valuable
knowledge management capability as 92% of the respondents
agree or strongly agree that this would support the exchange of
customer related knowledge in the organization. In addition,
68% of the participants also estimate that CI Analytics would
support the coordination of the customer related activities.

Figure 6.9.: CI Analytics: Business Benefit 2 – Question 7

206

6. CI Analytics Software

Figure 6.10.: CI Analytics: Business Benefit 2 – Question 8
• Business Benefit 3: CI Analytics allows an analysis over time
and a benchmarking of the relationships established with
customers and supports the provider’s decisions concerning
investments in customers.
This third business benefit is evaluated with the following two
items:
– Question 9: Analyzing the evolution of this relationship
network overview over time would help us monitoring the
relationship with the customer.
– Question 10: Together with other indicators such as sales
results, this information would help us compare the performance achieved with different customers and would
help us in our choice to invest in one or the other customer.
The results of the assessment of these two questions is outlined in figures 6.11 and 6.12. The answers to question 9 show
that the survey participants overall believe in the ability of the
software to monitor customer relationships, even though the
results are less pronounced than for the previous items. The

6.3. CI Analytics Evaluation

207

answers to question 10 demonstrate that some of the respondents question the ability of the software to assess the performance achieved with different customers. This aspect may be
explained by the fact that our research on the leveraged customer intimacy developed in chapter 5 was not complete at the
time of the survey.

Figure 6.11.: CI Analytics: Business Benefit 3 – Question 9

Figure 6.12.: CI Analytics: Business Benefit 3 – Question 10
In conclusion, the following two items were assessed by the participants in order to capture their overall appreciation of the CI Analytics

208

6. CI Analytics Software

prototype and to determine the importance of data privacy issues related to this project:
• Question 11: I think such a visualization would be useful in our
company.
• Question 12: I would have privacy concerns if this type of information was made available in my company.

Figure 6.13.: CI Analytics: Overall Appreciation and Data Privacy
Concerns

Figure 6.14.: CI Analytics: Overall Appreciation and Data Privacy
Concerns
Figures 6.13 and 6.14 presents the results obtained for these two
items. The answers to question 11 confirm the relevance of the ap-

6.3. CI Analytics Evaluation

209

proach proposed by this thesis as 80% of the respondents agree or
strongly agree that such a visualization would be useful. In addition, the answers to question 12 demonstrate that even though data
privacy issues should be thoroughly addressed by a company adopting the software CI Analytics, they do not seem to a a strong obstacle
to the acceptance of the software by the employees as only 32% of
the respondents agree or strongly agree that they would have privacy concerns if this type of information was made available in the
company.
This chapter proved the feasibility of the calculation of the acquired
customer intimacy metrics at the organizational and individual levels as well as of the calculation of the leveraged customer intimacy
metrics through the conception and implementation of the software
CI Analytics. Moreover, a business benefits survey confirmed that
professionals involved in B2B activities would have a strong interest in such an application if it was available in their organization.
The next chapter will demonstrate the relevance of the CI Analytics
methodology proposed in chapter 5 for calibrating of the customer
intimacy metrics and, thereby, accurately assessing the customer intimacy components.

7. CI Analytics Validation
In order to perform the assessment of the degree of customer intimacy established by a provider with his different customers in a B2B
context, this thesis elaborated in chapter 5 the CI Analytics model and
methodology. As depicted in figure 5.1, the CI Analytics methodology
consists of seven steps. The first three steps concern the breakdown
analysis of customer intimacy in multiple quantifiable components,
the identification of data sources holding evidence of customer intimacy, and the determination of metrics to assess the customer intimacy components upon this data. Chapters 4 and 5 detailed the
completion of these three steps and provided, thereby, the foundations for the assessment of customer intimacy.
The steps 4 to 7 of the CI Analytics methodology, as explained in
section 5.1, support the identification of the most relevant metrics to
perform an accurate inference of the customer intimacy components
as well as allow a consideration of the specific activity and interaction patterns of the provider in the determination of the relative
importance of the customer intimacy metrics. Step 4 refers to the
actual calculation of the metrics for a specific customer. Step 5 concerns the empirical assessment of the customer intimacy components
by means of a survey with provider employees. Step 6 relates to the
application of machine learning algorithms on the metrics calculated
in step 4 in order to predict the empirical results obtained in step 5.

212

7. CI Analytics Validation

Step 6, thus, results in a set of machine learning models which contain information about the most relevant metrics to accurately infer
the customer intimacy components. Finally, step 7 refers to the validation and interpretation of the created machine learning models in
order to derive some managerial implications. In this chapter, this
sequence from step 4 to step 7 which is individually performed for
each provider is referred to as the calibration of the customer intimacy
metrics.
This chapter will show how this calibration has been performed in
a real-case scenario and will evaluate to which extent the customer
intimacy components acquired knowledge of, and established relationships with, customers have been inferred from the customer intimacy
metrics. This chapter will, thus, validate the overall approach taken
by this thesis to assessing and monitoring customer intimacy in a
B2B context. This validation has been performed with the support of
the IT software and services provider CAS Software AG (CAS). The
customer intimacy metrics have been calculated for 14 different CAS
customers upon the data stored in CAS genesisWorld. In addition, 25
CAS employees performed the empirical assessment of the customer
intimacy components for these 14 customers.
The CI Analytics model developed in figure 5.2 establishes that the acquired knowledge of, and established relationships with, customers1
should be assessed at two levels of detail: the individual level and
the organizational level. Consequently, the calibration of the customer intimacy metrics has been performed four times in order to
determine the best metrics to infer the values of these two components at these two levels of detail. Section 7.1 will elaborate on
the results of the calibrations of the customer intimacy metrics performed to predict the values of the component acquired knowledge
and established relationships at the individual level. Section 7.2 will
subsequently develop the results of the calibrations of the customer
intimacy metrics to predict these components at the organizational
level.
1

These components are called acquired knowledge and established
relationships in the remaining of this chapter.

7.1. Acquired Customer Intimacy at the Individual Level

213

7.1. Acquired Customer Intimacy at the
Individual Level
This section presents the results of the calibration of the customer intimacy metrics to assess the acquired customer intimacy components
acquired knowledge and established relationships at the individual
level. Part 7.1.1 describes the data collection process which corresponds to the calculation of the customer intimacy metrics and to the
empirical assessment of the customer intimacy components. Subsequently, parts 7.1.2 and 7.1.3 present the results of the application
of machine learning algorithms on the calculated customer intimacy
metrics to infer the values of the components acquired knowledge
and established relationships.

7.1.1. Data Collection
This section consists of two parts. Part 7.1.1.1 details the setup of
the calculation of the customer intimacy metrics at the individual
level for 14 CAS customers. This corresponds to the step 4 of the
CI Analytics methodology. Subsequently, part 7.1.1.2 elaborates on
the survey performed to assess acquired knowledge and established
relationships at the individual level for these 14 customers. This
activity refers to the step 5 of the CI Analytics methodology.
7.1.1.1. Calculation of the Customer Intimacy Metrics
It is explained in section 5.2.2 that, given a certain set of parameters,
eight metrics can be calculated upon the concepts of customer interaction time and weighted customer interaction time. These metrics are
volume, weighted volume, intensity, weighted intensity, frequency, duration, number of episodes, and mode of interaction. The parameters are
the time period T, the segment duration d, the default customer interaction time values of emails demail and letters dletter , and finally the
three threshold parameters interaction duration threshold ∆, interaction
quantity threshold b, and weighted interaction quantity threshold wb. The
parameters values for calculating the customer intimacy metrics in
this scenario are summarized in table 7.1. They have been chosen
upon the following considerations:

214

7. CI Analytics Validation

Table 7.1.: Model Configurations and Metrics to Assess Acquired
Customer Intimacy at the Individual Level
Configuration

A

B

C

D

Time Period T

3 months

12 months

12 months

Segment Size d
Email CIT Value
demail
Letter CIT Value
dletter
Interaction Duration
Threshold ∆
Interaction Quantity
Threshold b
weighted Interaction
Quantity Threshold
wb

1 month
10
minutes
10
minutes

1 month
10
minutes
10
minutes

3 months
10
minutes
10
minutes

Over one
year
N/A
10
minutes
10
minutes

1 month

1 month

1 month

N/A

0

0

0

N/A

0

0

0

N/A

Volume
3M
Volume
Weighted
3M
Intensity
3M
Intensity
Weighted
3M
Frequency
3M
Duration
3M
Number
of
Episodes
3M

Volume
12M
Volume
Weighted
12M
Intensity
12M
Intensity
Weighted
12M
Frequency
12M
Duration
12M
Number
of
Episodes
12M
Mode
12M

Metrics
volume
weighted volume
intensity
weighted intensity
frequency
duration

number of episodes

mode of interaction

Mode 3M

Volume
More 1Y
Volume
Weighted
More 1Y

Frequency
Quarter

7.1. Acquired Customer Intimacy at the Individual Level

215

• Time Period T
Within the scope of this thesis, three time periods mapped to
the operational pace of the provider organization have been
considered:
– the first time period is set to three months. It reflects the
recent interactions that occurred in the past quarter and
potentially provides the newest updates on the customer
and his needs.
– the second time period is set to 12 months. This time period corresponds to the longer projects that occurred with
the customer during the past year.
– the third considered time period consists of all interactions
that occurred with the customer in the past, with the exceptions of the interactions that happened within the last
12 months. It is denoted as over one year. This time period
reflects the fact that some employees may have established
qualitative relationships with customer employees in the
past, even though they had no contact within the last year.
• Segment Size d
The segment size has to be specified in order to determine the
level of detail of the analysis. For the 3-month and 12-month
time periods, the main segment size is set to one month as this
level of analysis should provide well interpretable results. In
addition, the metric frequency is also calculated for the time
period 12 months with a segment size of three months in order
to gain further insights on the interaction regularity over the
past year. With regard to the time period over one year which
considers all interactions stored in the provider’s information
system between the beginning of the relationship with the customer and a year before the analysis is performed, a breakdown
of the calculation of the customer interaction time and weighted
customer interaction time in multiple time segments was technically not feasible with the prototype CI Graph. Thus, only the
metrics volume and weighted volume have been calculated for this
time period.

216

7. CI Analytics Validation

• Email and Letter Customer Interaction Time Values demail and
dletter
The parameters demail and dletter provide the means to convert
emails and letters exchanged with customers into customer interaction time values, thereby enabling the integration of emails
and letters in the calculation of the customer intimacy metrics.
In the context of this scenario, only the emails and letters exchanged with customers and containing some content which is
relevant for the other provider employees are stored in the application CAS genesisWorld. Thus, both demail and dletter have
been set in the context of this thesis to the value 10 minutes.
This value reflects the average time spent by provider employees to write and read such emails and letters. Future research
should investigate how to adjust the values of demail and dletter ,
taking for instance into account criteria such as the length of
the email or letter, or the roles of the involved employees.
• Threshold Parameters ∆, b, and wb
Finally, the different threshold parameters have to be specified.
As explained in section 5.2.2.1, the interaction duration threshold
∆ is set to one month, meaning that if no interaction occurs
within one segment, a new episode starts with the next segment
containing some interaction. The interaction quantity threshold b
and weighted interaction threshold wb are both set to their default
value 0 in order to capture and consider all interactions in the
calculation of the metrics.
Table 7.1 presents an overview of the four instantiated parameter
configurations as well as of the 19 resulting customer intimacy metrics. These 19 metrics have been calculated for all couples { p; c}
where p represents a CAS employee, c represents an employee of
one of the 14 considered customer organizations, and existing data
reveals that some interaction occurred in the past between p and c.
This calculation has been performed with the software CI Graph presented in appendix E.3 and resulted in a data set of 10077 records.
This data set is called the customer intimacy metrics data set. Each
record in this data set consists of a reference to a CAS employee, a

7.1. Acquired Customer Intimacy at the Individual Level

217

reference to a customer employee, and the values of the 19 customer
intimacy metrics.
7.1.1.2. Empirical Assessment
This activity corresponds to the step 5 of the CI Analytics methodology proposed in section 5.1.1. It refers to the empirical assessment
of the customer intimacy components by means of a survey with
provider employees. At the individual level, this assessment consists
of an evaluation by provider employees of their knowledge of, and
relationship with, customer employees. It is performed using 7-point
Likert-type scales with the following four Likert items. These items
are inspired from past literature and their selection is developed in
section 5.2.4.2
• Acquired knowledge of customer employees
– Item 2.1: “My knowledge of [CustomerEmployeeName]’s
needs is thorough.”
– Item 2.2: “I learned a lot about [CustomerEmployeeName]’s
preferences in the period I worked with him/her.”
• Established relationships with customer employees
– Item 2.3: “I have a high-quality relationship with [CustomerEmployeeName].”
– Item 2.4: “I have a very collaborative relationship with
[CustomerEmployeeName].”
Each provider employee participating in the empirical assessment
of the acquired customer intimacy components evaluates his knowledge of, and relationship with, different customer employees using these four Likert items. Thus, each provider employees answers
these four items multiple times, each time for a different customer
employee. Appendix A figure A.3 illustrates such a questionnaire
in which the survey participant is asked to assess his knowledge of,
2

Likert-type scales are explained in section 3.1.3.

218

7. CI Analytics Validation

and relationship with, seven different customer employees. Each of
these empirical assessments corresponds to a couple { pi ; c j } where pi
represents a CAS employee and c j represents a customer employee.
Thus, as depicted in figure 7.1, each empirical assessment can be associated with a record of the customer intimacy metrics data set proposed in section 7.1.1.1 which contains the 19 calculated customer
intimacy metrics.
Customer Intimacy Metrics Data Set

Empirical Assessment of the Customer
Intimacy Components (Survey)

p1

c1

Values of the 19 customer intimacy
metrics for the couple (p1, c1)

p1

c1

Results of the assessment by p1
of the four Likert-items for c1

pi

cj

Values of the 19 customer intimacy
metrics for the couple (pi, cj)

pi

cj

Results of the assessment by pi
of the four Likert-items for cj

...

... ...

...

...

...

Resulting Calibration Data Set
(prediction variables)

(underlying data of the predicted
variables)

p1

c1

Values of the 19 customer intimacy Results of the assessment by p1 of
metrics for the couple (p1, c1)
the 4 Likert-items for c1

pi

cj

Values of the 19 customer intimacy Results of the assessment by pi of
metrics for the couple (pi, cj)
the four Likert-items for cj

...

...

...

...

Figure 7.1.: Creation of the Calibration Data Set
43 CAS employees were proposed by CAS to participate in the empirical evaluation, with the constraint that each employee performs
a maximum of six assessments in order to limit their time investment. This means that, in total, a maximum of 258 assessments can
be performed. Thus, only 258 out of the 10077 records in the customer intimacy metrics data set can be associated with an empirical
assessment. A thorough sampling of the customer intimacy metrics
data set is, therefore, necessary in order to select these 258 records.
The purposeful sampling methodology is applied in this thesis in
order to manage this constraint on the sample size and select these
records. Purposeful sampling refers to the “selection of information-

7.1. Acquired Customer Intimacy at the Individual Level

219

rich cases for study in-depth” (Patton, 2002, p.45). These informationrich cases are those which have a high relevance for the purpose of
the investigation. Berry & Linoff (2004, p.63) confirm that “a smaller,
balanced sample is preferable to a larger one with a very low proportion of rare outcomes.” In the context of this thesis, the purpose of
the analysis is the assessment of the customer intimacy components
and, more specifically, the identification of the provider employees
that have gathered significant knowledge of, and established good
relationship with, customer employees. In the customer intimacy
metrics data set, several records have very low customer intimacy
metrics values, thereby indicating that very few interactions occurred
between the corresponding provider and customer employees. These
specific records are unlikely to be correlated with high degrees of
knowledge and relationship and, therefore, should be ignored as
they do not have a high relevance for our analysis. Three clusters
that reflect some relevant interaction patterns between the 43 surveyed employees and the customer employees have been considered
in order to create the sample:
• The first cluster contains records indicating that, over the past
year, the quantity of interaction was above 2.6 hours and some
face-to-face interaction occurred. These records have a high
probability of being correlated to high customer intimacy values, as people have met at least once in person. The records
pertaining to this cluster, therefore, fulfill the following two
conditions: Volume 12M > 2.6 and Mode 12M > 0. This cluster
contains in total 141 records.
• The second cluster contains records indicating that, over the
past year, the quantity of interaction was above 1 hour, but
no face-to-face interaction occurred. These records provide the
ability to evaluate the influence of face-to-face interaction on
the customer intimacy components. The records pertaining to
this cluster fulfill the following two conditions: Volume 12M >
1 and Mode 12M = 0. This cluster contains in total 54 records.
• Finally, in order to assess the impact of the interaction that
occurred before the past year, the third cluster contains the

220

7. CI Analytics Validation

records indicating that no interaction occurred within the last
year, but the customer interaction time before the past year is
above 5 hours. The records pertaining to this cluster fulfill the
following two conditions: Volume 12M = 0 and Volume More1Y >
5. This cluster contains in total 73 records.
Combining the three clusters, the overall sample contains a total of
232 records. It has been used in order to generate the questionnaires
of the 43 CAS employees participating in the empirical customer intimacy assessment: the CAS employees are asked to assess their knowledge of, and relationship with, provider employees which are referenced in these 232 records. Thus, each respondent receives a unique
questionnaire which is tailored to his past interaction with customer
employees. A custom application to generate them automatically
has been implemented in order to create these questionnaires in an
efficient manner.
Table 7.2.: Creation of the Calibration Data Set
Cluster

Conditions

Requested

Received

1

Volume 12M > 2.6 and Mode 12M > 0

141

50

2

Volume 12M > 1 and Mode 12M = 0

54

30

3

Volume 12M = 0 and Volume More1Y > 5

73

37

Total

232

127

The survey was conducted between October and November 2010.
25 out of the 43 employees returned their questionnaires, resulting
in 127 empirical assessments of the customer intimacy components.
Table 7.2 summarizes the outcome of the survey.
As illustrated in figure 7.1, in order to perform the calibration, the
results of the empirical assessment of the customer intimacy components by the provider employees are appended to the corresponding
records of the customer intimacy metrics data set and, thus, associated to the 19 calculated customer intimacy metrics. The resulting
calibration data set used for the determining the most relevant metrics

7.1. Acquired Customer Intimacy at the Individual Level

221

to assess the values of the customer intimacy components contains
127 records consisting of (i) the reference to a provider employee
(pi ); (ii) the reference to a customer employee (c j ); (iii) the 19 corresponding customer intimacy metrics values; and (iv) the empirical
assessment of the acquired customer intimacy components based on
the four previously introduced Likert items.

7.1.2. Calibration: Acquired Knowledge
This section refers to the steps 6 and 7 of the CI Analytics methodology proposed in section 5.1. It describes the calibration of the
customer intimacy metrics to determine the values of the component acquired knowledge at the individual level upon the customer intimacy metrics. This calibration consists of the application of machine
learning algorithms to learn how to infer the empirically assessed acquired knowledge values, as well as the validation and interpretation
of the resulting machine learning models. The pre-processing and
data transformation tasks are presented in parts 7.1.2.1 and 7.1.2.2.
Then, the application of machine learning algorithms as well as the
validation and interpretation of the machine learning models are described in part 7.1.2.3 and part 7.1.2.4.
7.1.2.1. Preprocessing
The pre-processing activity consists of three different tasks:
• Anonymize the Data Set
Since the respondents provide in their questionnaires some sensitive information about their relationships with different customer employees, the anonymity of the records in the data set
has to be strictly preserved. Moreover, in the scope of this thesis, information related to the characteristics of the individual
employees such as their role and position is not considered:
the calibration is based exclusively on the 19 metrics. Therefore, the references to the provider and customer employees in
all records of the data set are removed.
• Manage Missing Values
The missing values in the context of this project are twofold:

222

7. CI Analytics Validation

– first, they consist of the interactions and activities that
occurred between the provider and customer employees
which are not recorded in the information system. The
calculation of the customer intimacy metrics may be incorrect if such interaction data is missing. Since most of
the interaction data is automatically stored in CAS genesisWorld, it can be assumed that this type of missing value
is not significant and, thus, no specific corrective action is
undertaken. This aspect, though, should be considered in
the results interpretation.
– the second type of missing value refers to the Likert items
which have not been empirically assessed by the survey
respondents. In particular, this refers to the Likert items
2.1 and 2.2 developed in section 7.1.1.2. The respondents
did not assess both items 2.1 and 2.2 in 10 out of the 127
records. These 10 records are, therefore, removed from the
data set because they cannot be used for the calibration.
In addition, the respondents answered only one of the two
items in 16 records. The method “Imputation by Using Replacement Values” proposed by Hair et al. (2010, p.52) is
used in order to manage these missing values: this “form
of imputation involves replacing missing values with estimated values based on other information available in the
sample.” In this context, the value of the assessed item is
used to determine the value of the missing item. For instance, if the item 2.1 is answered with the value 1 and the
item 2.2 is missing, then the value of the item 2.2 is also set
to the value 1. As a result, the calibration data set consists
of 117 records.
• Manage Outliers
Outliers are “observations with a unique combination of characteristics identifiable as distinctly different from the other observations” (Hair et al., 2010, p.64). Such records may be significant as they can have a strong influence on the results of
analysis. These records may be deleted, transformed, or simply
kept unmodified in the data set. In this project, the third option

7.1. Acquired Customer Intimacy at the Individual Level

223

is chosen and the outliers are considered as any other record
for three reasons. First, these outliers cannot be considered as
noise. These records may represent some valid patterns of interactions, even though different from others and, thus, should
be considered in the training phase of the machine learning
activity. Second, the objective of this calibration is to create a
machine learning model that can be used to assess the customer
intimacy components out of the customer intimacy metrics. If
the outliers are transformed or removed from this specific data
set, the machine learning model might not be accurate when
applied to other data sets where the outliers are not removed
or transformed. Finally, some of the chosen machine learning
algorithms presented in section 3.2.2 are “robust” or “resistant
to outliers” (Tan et al., 2006, p.38). For instance, the decision
tree C4.5 includes a pruning option in order to limit the outliers
influence on the design of the machine learning model (John,
1995, p.1).

7.1.2.2. Data Transformation
The data transformation task concerns the aggregation of the two
items 2.1 and 2.2 presented in section 7.1.1.2 in order to create the
required target value to apply the supervised learning approach, as
explained in section 3.2.1. This transformation is performed in two
steps:
• Creation of the Summated Scale Knowledge
First, a summated scale “formed by combining several individual variables into a single composite measure” is created (Hair
et al., 2010, p.124). The proposed summated scale is denoted
Knowledge and is calculated as the mean of V(Item 2.1) and
V(Item 2.2) which represent the empirically assessed values of
the items 2.1 and 2.2 :
Knowledge =

V(Item 2.1) + V(Item 2.2)
2

(7.1)

224

7. CI Analytics Validation

Two requirements have to be considered in the creation of the
summated scale. First, the content of the summated scale has
to be conceptually valid and its components should represent
the same dimension. In this project, a top-down approach has
been used to create the scale, and the items of the questionnaires have been determined upon existing scales presented in
past literature, as explained in chapter 4. Thus, the conceptual validity of the scale is ensured. Second, the reliability of
the scale should be verified. Reliability is an “assessment of
the degree of consistency between multiple measurements of
a variable” (Hair et al., 2010, p.125). The Crombach’s Alpha
test is performed on the data set in order to verify the reliability of the proposed summated scale. If this test returns a value
above 0.70, the summated scale is considered as reliable (Robinson et al., 1991). As presented in Appendix C figure C.1, the
Crombach’s Alpha test on the scale Knowledge returned the high
value of 0.912. Thus, this scale is conceptually valid and reliable.
• Knowledge Scale Binarization
The previously created scale Knowledge consists of 13 categories
ordered from the value 1 to the value 7 by increments of 0.5:
{1, 1.5, 2, ..., 6, 6.5, 7}. As explained in section 3.2.2, since the
Likert-type scales used in the questionnaires are considered as
ordinal, the scale Knowledge is also ordinal. Thus, the purpose
of the calibration is to create a machine learning model capable of predicting the class of each record in the sample. The
binarization method presented in Witten et al. (2011, p.315) is
applied in this project on the Knowledge scale. This binarization
method converts the 13-class classification task into multiple
2-class classification tasks. The reason for the wide adoption
of this technique in data mining projects is that many machine
learning algorithms perform better or even are only applicable
on binary classification problems (Witten et al., 2011, p.315).
The creation of a binary classification task for each class of the
Knowledge scale would result in 13 classification tasks. For instance, a binary variable would be set to 1 if the record belongs

7.1. Acquired Customer Intimacy at the Individual Level

225

to the class is 1, and to 0 otherwise. Another one would be set to
1 if the record belongs to the class 1.5, and to 0 otherwise. Such
a level of detail is, however, not required in this thesis: from
a business perspective, the objective is to assess whether the
provider employees have no knowledge, a high knowledge or
a very high knowledge of specific customer employees. Thus,
two binary indicators have been created:
– Knowledge High: This variable is designed to identify the
records indicating that a provider employee has acquired
some knowledge of a customer employee. The limit to
consider that a provider employee has some knowledge
about a customer employee is set to the median value of
the Knowledge scale which is equal to 4. Thus, the Knowledge High variable is set to 1 if the variable Knowledge is
equal or above 4.5, and it is set to 0 otherwise:

Knowledge High =

if Knowledge ≥ 4.5
otherwise

1
0

– Knowledge Very High: This variable serves the identification of records indicating that a provider employee has a
very high knowledge of a customer employees. It is considered in this thesis that a provider employee estimates its
knowledge of a provider employee as very high if he answers the items 2.1 and 2.2 of the questionnaires with values above 6. Thus, the Knowledge Very High variable is set
to 1 if the variable Knowledge is equal or above 6, and it is
set to 0 otherwise:

Knowledge Very High =

1
0

if Knowledge ≥ 6
otherwise

As described in table 7.3, within the data set consisting of 117 records,
the proportions of records in which the variables Knowledge High and

226

7. CI Analytics Validation

Table 7.3.: Proportions of Knowledge High and Knowledge Very High
Records
Quantity of
Records with
Value 1

Quantity of
Records with
Value 0

Total
Quantity of
Records

Knowledge
High

56 (48%)

61 (52%)

117 (100%)

Knowledge
Very High

35 (30%)

82 (70%)

117 (100%)

Knowledge Very High are set to the value 1 are 48% and 30%. Logically,
there are fewer records in which the respondent estimated having a
very high knowledge than having a high knowledge of the customer
employee.
The next subsections 7.1.2.3 and 7.1.2.4 focus on the creation of machine learning models in order to predict whether the records in the
sample belong to the classes Knowledge High and Knowledge Very High.
7.1.2.3. Knowledge High Calibration and Validation
In this section, the results of customer intimacy metrics calibration
to predict the value of the variable Knowledge High are presented.
The method “10-times 10-fold cross-validation” which is explained
in section 3.2.3 is applied in order to jointly create the machine learning models and validate their performance. In addition, section 3.2.3
describes the indicators used in this project to assess the performance
of the resulting machine learning models. These performance indicators are the following: success rate, precision, recall, F-measure, and
kappa statistic.
These performance indicators have to be considered in the context
of the project in order to be interpreted. A precision of 70% may be
considered as low in a certain project and high in another one, depending of the project objectives, results implications, and quality of
the data set. In order to facilitate this interpretation, three intervals

7.1. Acquired Customer Intimacy at the Individual Level

227

which are denoted as good, fair, and poor are determined for each performance indicator. All calibration results which are presented in the
next sections are, thus, clustered along these three intervals. However, to ensure the completeness of this thesis, the actual key performance indicator values are also detailed for all calibration results.
Table 7.4 summarizes the interval values for the five performance
indicators. These intervals are determined upon on the following
considerations:
• Precision
In this project, precision is considered as the most important indicator. Assuming that an organization considers the adoption
of the CI Analytics model and methodology and the deployment
of the software CI Analytics presented in chapter 6, this organization will expect precise and reliable results. Considering the
size and quality of the data set, the precision is defined as good
if it is above 80%, fair if it is between 60% and 80%, and poor
otherwise. A precision above 80% indicates that at least four
out of five records predicted as “Knowledge High” are actually
of class “Knowledge High”.
• Recall
Recall is considered in this project as less important than precision because the machine learning models could easily be complemented later with additional customer intimacy metrics in
order to improve the capability of the model to retrieve the
records of class “Knowledge High”. Thus, recall is considered as
good if is is above 70%, fair if it is between 50% and 70%, and
poor otherwise.
• Success Rate
The success rate represents the overall ability of the machine
learning models to predict the class of a record, regardless of
its actual class. In this project, the success rate is estimated as
good if it is above 75%, fair it is between 60% and 75%, and
poor otherwise.
• F-Measure
The F-measure, as explained in section 3.2.3, is a combination of

228

7. CI Analytics Validation

the precision and recall indicators, calculated as their geometric
mean. Thus, the intervals good, fair, and poor of the F-Measure
are also derived from the geometric means of the recall and
precision indicators. The F-Measure is considered as good if it
is above 75%, fair if it is between and 55% and 75%, and poor
otherwise.
• Kappa Statistic
The Kappa statistics compares the success rate of the machine
learning algorithm with the success rate achieved by a random
prediction. It is assumed that the Kappa statistic value is considered as good if the model is at least 50% better than the random predictor, fair if it is 40% to 50% better than the random
predictor, and poor otherwise.
Table 7.4.: Proposed Interpretation of the Performance Indicators
Interpretation
Good (%)

Fair (%)

Poor (%)

Precision

[80 − 100] [60 − 80[

[0 − 60[

Recall

[70 − 100] [50 − 70[

[0 − 50[

Success Rate

[75 − 100] [60 − 75[

[0 − 60[

F-Measure

[75 − 100] [55 − 75]

[0 − 55[

Kappa Statistic

[50 − 100] [40 − 50[

[0 − 40[

The four machine learning algorithms used to perform the calibration in order to predict the variable Knowledge High are the following: decision tree C4.5, k-nearest neighbor algorithm, support vector
machine algorithm, and multilayer perceptron with backpropagation
neural network. These algorithms are described in section 3.2.2. The
data-mining application Weka3 is used in order to perform the calibration.
3

Further information on Weka is available at
http://www.cs.waikato.ac.nz/ml/weka/ (accessed on 23.10.2011).

7.1. Acquired Customer Intimacy at the Individual Level

229

In order to optimize the performance of the different machine learning algorithms, the algorithms were not only trained and tested with
their default settings, but several parameter configurations were evaluated. Table C.1 in appendix C illustrates the series of configurations
considered for the optimization of the decision tree C4.5. It can be observed in this table that over 50 different configurations were tested,
each of them optimizing one of the parameters. The list of parameters, their descriptions as well as the parameter values considered
in this thesis are detailed in appendix B. Overall, most of the configurations presented in table C.1 led to fair or good results. The
model number 40 has been selected as it presents the best combination of precision and recall values (84% and 70%). Further information on these “best results” configurations for the decision tree C4.5
and for the other three machine learning algorithms is presented in
table C.2. The details on each tested configuration performed with
the k-nearest neighbor, support vector machine, and multilayer perceptron neural network algorithms are not presented in this thesis
but are available upon request from the author.
Table 7.5 presents the best results achieved with each of the four
machine learning algorithms. Overall, according to the interpretation
intervals proposed in table 7.4, all algorithms achieve good results
to predict the value of the variable Knowledge High. The decision
tree C4.5 and multilayer perceptron neural network obtained the best
results and achieved the grade good for all five indicators. On the
other hand, the k-nearest neighbor and the support vector machine
algorithms only obtained a fair recall value of 67.0%.
The “Receiver Operational Characteristic” (ROC) curve4 of the model
created with decision tree C4.5 is illustrated in figure 7.2(b). It shows
that this algorithm is very efficient in order to identify the first 72%
of the true positive records as the corresponding false positive ratio
remains below 10%. The size of the “area under ROC” is also high
with a value of 82%.
The best model created with the decision tree C4.5 algorithm is presented in figure 7.2(a). The first criteria of the tree considers the
4

Further details on the ROC curve are provided in section 3.2.3.

230

7. CI Analytics Validation

Table 7.5.: Prediction of the Variable Knowledge High: Performance
Indicator Results (g=good; f=fair; p=poor)
Model

Precision

Recall

Success Rate F-measure

Kappa

(%)

(%)

(%)

(%)

(%)

C4.5

84.0 (g)

70.0 (g)

78.4 (g)

75.0 (g)

56.0 (g)

k-NN

83.0 (g)

67.0 (f)

77.1 (g)

72.0 (f)

54.0 (g)

SVM

87.0 (g)

67.0 (f)

79.1 (g)

74.0 (f)

58.0 (g)

NNBP

87.0 (g)

71.0 (g)

80.2 (g)

76.0 (g)

60.0 (g)

metric Frequency 12M. This indicates that interaction regularity is
significant in order to obtain a high knowledge of a customer employee. More specifically, the criteria Frequency 12M > 25% can be
interpreted as follows: if the provider employee interacted with the
customer employee in at least four different months over the past
year, then he considers to have a high knowledge of the customer
employee.
The second criteria of the decision tree uses the metric Volume More1Y.
If the value of the metric Volume More1Y is below 1.2 hours, indicating that the provider employee interacted less than 1.2 hours with the
customer employee before the past year, then the provider employee
does not have a high knowledge of the customer employee. The third
considered criteria of the decision tree is based on the metric Volume
Weighted 12M. This metric reflects the weighted customer interaction
time over the past year. The value of the metric Volume Weighted 12M
has to be above 0.375 hour so that the provider employee considers to have a high knowledge of the customer employee. Since the
weighted metrics take the number of participating employees to each
interaction into consideration, this criteria can be achieved in multiple ways: for instance, the provider employee can have a meeting
of 0.375 hour alone with the customer employee, or he can meet the
customer employee together with three other persons for a duration
above 1.5 hours (0.375 × 4).

7.1. Acquired Customer Intimacy at the Individual Level

231

100

Frequency 12M
90

> 25%

Volume More1Y
≤ 1.2

True (40.0 / 4.0)
> 1.2

Volume Weighted
12M

False (30.0 / 2.0)

80

True Positive (%)

≤ 25%

70
60
50

Area under ROC: 0.82
40
30

20

≤ 0.375

> 0.375
10

False (41.0 / 12.0)

True (6.0)

0
0

20

40

60

80

100

False Positive (%)

(a) Decision Tree Representation

(b) ROC Curve
Figure 25 (2) ROC Curve C4.5 Indiv Knowledge High

High Decision
Tree Decision Tree Model and ROC Curve
FigureFigure
7.2.:24 Knowledge
Knowledge
High:

The following managerial implication can be derived from these results: a company willing to foster the acquisition of knowledge related to customer employees should encourage its own employees
to regularly interact with the customer. More specifically, the provider employees should interact with the customer employees in at
least four different months every year (Frequency 12M > 25%). These
results also indicate that employees who interacted with customer
employees in the past, but not within the last year still have a good
knowledge of these customer employees and could be contacted if
such knowledge was required in the organization.
7.1.2.4. Knowledge Very High Calibration and Validation
The four machine learning algorithms decision tree C4.5, k-nearest
neighbor, support vector machine, and multilayer perceptron neural
network have been trained and tested in multiple configurations in
order to optimize the prediction of the variable Knowledge Very High.
These different series of configurations are available upon request
from the author. The configurations that lead to the best results with
each of the algorithms are described in appendix C table C.3.
Table 7.6 summarizes the best results achieved with the four algorithms. In order to interpret these performance indicators as good,
fair or poor, the intervals determined to predict the variable Knowledge High which are described in table 7.4 have been used. Since

232

7. CI Analytics Validation

the proportion of records of class Knowledge Very High is significantly
lower than the proportion of records of class Knowledge High (30% vs.
48%, see table 7.3), retrieving the records of class Knowledge Very High
is more difficult for the machine learning algorithms than retrieving
the records of class Knowledge High. Thus, lower precision and recall
values are to be expected.
While the results achieved to predict the variable Knowledge High
were homogeneous with the four algorithms, the results achieved
with regard to the prediction of the variable Knowledge Very High are
disparate. The multilayer perceptron neural network obtained worse
results than the other three algorithms according to all performance
indicators. Its recall and precision values are only equal to 61.0%
and 60.0%, meaning that the algorithm only retrieved 60.0% of the
records of class Knowledge Very High, and from all records predicted
as belonging to the class Knowledge Very High, only 61.0% of them
were correct. With regard to the performance indicator success rate,
the decision tree C4.5, the k-nearest neighbor, and support vector
machine algorithm all achieved good results above 79.0%. However,
the k-nearest neighbor is the only algorithm that achieved a good
precision with a value of 83.0%, but its recall value remains only fair,
with a value of 57.0%.
Table 7.6.: Knowledge Very High: Performance Indicator Results
(g=good; f=fair; p=poor)
Model

Precision

Recall

Success Rate

F-measure

Kappa

(%)

(%)

(%)

(%)

(%)

C4.5

72.0 (f)

55.0 (f)

79.4 (g)

59.0 (f)

46.0 (f)

k-NN

83.0 (g)

57.0 (f)

83.5 (g)

64.0 (f)

55.0 (g)

SVM

71.0 (f)

62.0 (f)

80.3 (g)

63.0 (f)

50.0 (g)

NNBP

61.0 (f)

60.0 (f)

76.5 (g)

58.0 (f)

42.0 (f)

The ROC curve of the best model obtained with the k-nearest neighbor algorithm is presented in figure 7.3(b). This model is highly

7.1. Acquired Customer Intimacy at the Individual Level

233

efficient in order to retrieve the first 50% of the records of class Knowledge Very High as the corresponding false positive ratio remains below 5%. Thus, if the objective is to identify some provider employees
who have a very high knowledge of specific customer employees,
this algorithm performs very well.
Figure 7.3(a) presents the decision tree created with the best configuration of the C4.5 algorithm. This model should be interpreted cautiously as it only achieves some fair results. The metric Intensity 12M
is considered in the first node of the tree (Intensity 12M > 1.353). This
indicates that the average interaction duration is an important aspect
in order to obtain a very high knowledge of a customer employee: if
the interaction of the provider employee with the customer employee
last on average over than 1.353 hours, then the provider employee
obtains a very high knowledge of the customer employee. If the
Intensity 12M is below 1.353 hours, the remaining criteria of the decision tree use regularity-based metrics, thereby indicating that some
regularity in the interaction is required in order for the provider employee to acquire a very high knowledge of the customer employee.
The second node of the tree is based on the metric Frequency Quarter and tests whether the provider and the customer employee interacted in at least two different quarters over the past 12 months
(Frequency Quarter > 25%). The third node (Number of Episodes 3M >
0) checks whether some interaction occurred within the last three
months. Finally, the fourth node Frequency 12M tests whether the
interaction was not too regular: if interaction occurred in more than
seven months (Frequency 12M > 58.33%), then the provider employee
does not have a very high knowledge of the customer employee. This
last aspect may be interpreted as follows: the provider employees
who are responsible for sending very regular information to customers, such as newsletters and advertisement do not have a very
high knowledge of the provider employees.
These results lead to the following managerial implications. First,
since most of the metrics used in the decision tree are based on the
regularity of the interactions, these results confirm the findings of
section 7.1.2.3 that a company should ensure that its employees regularly interact with customer employee in order to personally know

234

7. CI Analytics Validation

them. Second, since the first criteria of the tree based on the metric Intensity 12M, a company should organize customer events such
as workshops or consulting projects in which the provider and customer employees work together on a long period of time. Such interactions allow the provider employees to obtain a very high knowledge of the customer employees.
Intensity 12M

100
90

> 1.353

≤ 1.353

80

≤ 25%

True (11.0)

True Positive (%)

Frequency Quarter
> 25%

Number of Episodes
3M

False (81.0 / 10.0)

>0

≤ 0.0

False (6.0 / 1.0)

70
60
50

Area under ROC: 0.76
40
30
20

Frequency 12M

10

≤ 58.33

> 58.33%

0
0

True (12.0 / 2.0)

False (7.0 / 2.0)

(a) Decision Tree Representation

20

40

60

80

100

False Positive (%)

(b) k-nearest
Neighbor
Model ROC Curve

Figure 27 ROC Curve kNN Indiv Knowledge Very High

Knowledge Very High Decision Tree

Figure 7.3.: Knowledge Very High: Decision Tree Model and k-nearest
Neighbor ROC Curve

7.1.3. Calibration: Established Relationships
The results of the calibration and validation of the customer intimacy metrics in order to assess the customer intimacy component
established relationships at the individual level are presented in this
section. This corresponds to the steps 6 and 7 of the CI Analytics
methodology. Since these activities have already been thoroughly
described in section 7.1.2 for the assessment of the customer intimacy component acquired knowledge, this section mainly focuses on
the description of the results.
7.1.3.1. Preprocessing
As previously explained in section 7.1.2.1, the pre-processing activity
consists of three different tasks:

7.1. Acquired Customer Intimacy at the Individual Level

235

• Anonymize the Data Set
References to the provider and customer employees are removed
from the data set in order to ensure the anonymity of the analysis.
• Manage Missing Values
While no corrective action is undertaken in order to manage
the missing interaction records in the database, the missing
data related to the empirical assessment of the customer intimacy components is managed with the method presented in
section 7.1.2.1. The items 2.3 and 2.4 have been assessed by the
respondents in order to determine the value of the customer intimacy component established relationships. Within the original
sample of 127 records, 23 records do not contain an assessment
of items 2.3 and 2.4. Thus, these 23 records are removed. Then,
there is no record in which either the item 2.3 or the item 2.4
has been assessed. Therefore, the final data set for the calibration of the customer intimacy component established relationships
consists of 104 records.
• Manage Outliers
The outliers are kept unchanged in the data set, as explained in
section 7.1.2.1.
7.1.3.2. Data Transformation
As for the assessment of the customer intimacy component acquired
knowledge, the data transformation activity consists of two tasks: the
conception of the summated scale Relationship and its binarization
with the creation of the indices Relationship High and Relationship Very
High.
• Creation of the summated scale Relationship
The scale Relationship is calculated as the mean of V(Item 2.3)
and V(Item 2.4) which are the empirically assessed values of the
items 2.3 and 2.4. Similarly to the scale Knowledge presented in
section 7.1.2.2, the scale Relationship is conceptually valid as a
top down approach has been followed for its creation, and the

236

7. CI Analytics Validation

items 2.3 and 2.4 are derived from past literature. Moreover,
this scale is also reliable as its Crombach’s alpha value is equal
to 0.940, as illustrated in figure C.1.
Relationship =

V(Item 2.3) + V(Item 2.4)
2

(7.2)

• Relationship Scale Binarization
The scale Relationship consists of 13 classes that range from the
value 1 to the value 7 by increments of 0.5: {1, 1.5, ..., 7}. The binarization method is applied in order to transform the 13-class
classification task into two 2-class classification tasks. Thus, two
indices, Relationship High and Relationship Very High are created:
– Relationship High: The binary variable Relationship High
distinguishes the records indicating a high quality relationship from others. Similarly to the variable Knowledge
High, a relationship is considered as “high” if the Relationship value is above the median value of the Likert-scale.
Thus, the variable Relationship High is set to 1 if Relationship is equal of above 4.5, and to 0 otherwise. As described
in table 7.7, within the sample of 104 records, 59 records
belong to the class Relationship High, representing 56.7% of
the data set.

1
if Relationship ≥ 4.5
Relationship High =
0
otherwise

– Relationship Very High: in this project, a relationship between a provider employee and a customer employee is
considered as “very high” if the value of the corresponding record on the Relationship scale is equal or above 6. The
variable Relationship Very High is set to 1 if Relationship is
equal or above 6, and to 0 otherwise. 30 records in the
sample belong to the class Relationship Very High. They

7.1. Acquired Customer Intimacy at the Individual Level

237

represent a proportion of 28.8%, as illustrated in table 7.7.

1
if Relationship ≥ 6
Relationship Very High =
0
otherwise

Table 7.7.: Proportions of Records of Class Relationship High and Relationship Very High
Quantity of
Records with
Value 1

Quantity of
Records with
Value 0

Total
Quantity of
Records

Relationship
High

59 (56.7%)

45 (43.3%)

104 (100%)

Relationship
Very High

30 (28.8%)

74 (71.2%)

104 (100%)

The next subsections 7.1.3.3 and 7.1.3.4 present the results of the calibration of the customer intimacy metrics in order to predict the values of the variables Relationship High and Relationship Very High.
7.1.3.3. Relationship High Calibration and Validation
The four machine learning algorithms decision tree C4.5, k-nearest
neighbor, support vector machine, and multilayer perceptron with
backpropagation have been trained and tested with the 10 times 10fold crossvalidation methodology in order to calibrate the customer
intimacy metrics for the prediction of the variable Relationship High.
Several configurations were tested with each algorithm and are available upon request from the author. The best results achieved with
each algorithm as well as the corresponding configurations are described in appendix C table C.4.
Table 7.8 summarizes the performance of the four algorithms to predict the variable Relationship High. The algorithms perform overall
well, but slightly worse than to predict the variable Knowledge High.
The decision tree C4.5, the k-nearest neighbor, and the support vector

238

7. CI Analytics Validation

machine achieve a good precision over 80.0%. The multilayer perceptron achieves a precision just below the good limit, with a precision
of of 79.0%. The k-nearest neighbor and the support vector machine
perform better than the decision tree C4.5 as they also achieve higher
recall values of respectively 75.0% and 69.0%. The k-nearest neighbor is the only algorithm achieving both good precision and recall
values, at the cost of a fair overall success rate of 73.3%. This algorithm also obtains a good F-measure value of 76.0% and a fair Kappa
statistic value of 45.0%.
Table 7.8.: Relationship High: Performance Indicator Results (g=good;
f=fair; p=poor)
Model

Precision

Recall

Success Rate F-measure

Kappa

(%)

(%)

(%)

(%)

(%)

C4.5

80.0 (g)

59.0 (f)

67.0 (f)

65.0 (f)

35.0 (p)

k-NN

80.0 (g)

75.0 (g)

73.3 (f)

76.0 (g)

45.0 (f)

SVM

86.0 (g)

69.0 (f)

75.4 (g)

75.0 (g)

51.0 (g)

NNBP

79.0 (f)

72.0 (g)

70.9 (f)

73.0 (f)

41.0 (f)

The ROC curve of the best k-nearest neighbor configuration is illustrated in figure 7.4(b). This diagram indicates that the algorithm is
very efficient in order to retrieve the first 50% of the records of class
Relationship High, as the corresponding false positive rate remains below 10%. This algorithm, however, performs significantly worse in
order to retrieve the remaining 50% of true positive records.
Figure 7.4(a) depicts the tree created by the decision tree C4.5 algorithm. Interaction regularity followed by interaction quantity are the
two main aspects leading to a qualitative relationship with customer
employees. The first node of the tree considers the metric Frequency
Quarter. Provider employees consider having a high quality relationship with customer employees if they interacted with them in two
or more quarters over the past year (Frequency Quarter > 25%). If
the value of the metric Frequency Quarter is equal or below 25%, the
second criteria of the tree uses the metric Number of Episodes 12M.

7.1. Acquired Customer Intimacy at the Individual Level

239

100

Frequency Quarter

90

> 25%

Number of Episodes
12M
>1

≤1

False (61.0/21.0)

True (19.0)

Volume Weighted
3M
≤ 1.08

> 1.08

80

True Positive (%)

≤ 25%

70
60
50

Area under ROC: 0.75
40
30
20
10

True (7.0/1.0)

False (2.0)

0
0

20

40

60

80

100

False Positive (%)

(a) Decision Tree Representation

Figure 29 Relationship Very High C4.5 ROC Curve

(b) k-nearest Neighbor Model
ROC Curve

Figure 28 Relationship High Decision Tree

Figure 7.4.: Relationship High: Decision Tree Model and k-nearest
Neighbor ROC Curve

If there was no episode or only one episode of interaction over the
past year (Number of Episodes 12M ≤ 1), then the provider employees
do not consider having a qualitative relationship with the customer
employees. If there was more than one episode of interaction within
the last year, the third criteria of the tree is based on the metric Volume
Weighted 3M. This metric focuses on the interaction quantity over the
past three months: if all previous criteria are met and if the value
of the metric Volume Weighted 3M is below 1.08 hours, then the variable Relationship High is set to the value 1. This last criteria should,
however, be considered cautiously as it concerns a low number of
records.
From a managerial perspective, these results indicate that a regularity in the customer interaction is necessary in order for the provider employee to establish a qualitative relationship with customer
employees. Since the first criteria of the tree is based on the metric
Frequency Quarter, the regularity of the interaction over the past year
is particularly important and provider employees should meet customer employees in different quarters of the year in order to develop
qualitative relationships with them.

240

7. CI Analytics Validation

7.1.3.4. Relationship Very High Calibration and Validation
This section describes the results of the customer intimacy metrics
calibration for the prediction of the variable Relationship Very High.
The same four machine learning algorithms decision tree C4.5, knearest neighbor, support vector machine, and multilayer perceptron with back propagation neural network have been trained and
tested by means of 10 times 10-fold cross-validation on the 104-record
dataset.
Table 7.9 summarizes the best performance achieved with each algorithm. Further details on the corresponding parameter configurations are available in appendix C table C.5. As for the prediction
of the variable Knowledge Very High, the neural network algorithm
achieves poor results with a precision of 50.0% and a recall value of
51.0%. Even though the other three algorithms achieve good success
rates with values comprised between 77.4% and 81.1%, none of the
algorithm achieves a good precision in order to predict the value of
the variable Relationship Very High. The decision tree C4.5 algorithm
obtains the highest precision with a fair value of 75.0%. Its recall
values is also fair at 52.0%. This indicates that further metrics are
required in order to achieve a good performance on the prediction of
the variable Relationship Very High.
Table 7.9.: Relationship Very High: Performance Indicator Results
(g=good; f=fair; p=poor)
Model

Precision

Recall

Success Rate

F-measure

Kappa

(%)

(%)

(%)

(%)

(%)

C4.5

75.0 (f)

52.0 (f)

81.1 (g)

58.0 (f)

48.0 (f)

k-NN

66.0 (f)

55.0 (f)

77.8 (g)

57.0 (f)

43.0 (f)

SVM

65.0 (f)

52.0 (f)

77.4 (g)

54.0 (p)

41.0 (f)

NNBP

50.0 (p)

51.0 (f)

74.9 (f)

47.0 (p)

34.0 (p)

Figure 7.5(b) illustrates the ROC curve of the best model created with
the decision tree C4.5 algorithm. Similarly to the prediction of the

7.1. Acquired Customer Intimacy at the Individual Level

241

variable Relationship High, this diagram indicates that the algorithm
performs very well to identify the first 52% of records of class Relationship Very High, but it is inefficient to retrieve the remaining ones.
Frequency Quarter

100
90

> 50%

≤ 50%

False (83.0/14.0)

Mode 3M
>0

≤0

Volume Weighted
More1Y
≤ 9.936

True (7.0)
> 9.936

True Positive (%)

80
70
60
50

Area under ROC: 0.64
40
30
20
10

False (7.0/2.0)

True (7.0)

0
0

20

40

60

80

100

False Positive (%)

(a) Decision Tree Representation

(b) Decision Tree ROC Curve
Figure 31 Relationship Very High C4.5 ROC Curve

7.5.:
Relationship
FigureFigure
30 Relationship
Very
High Decision Tree

Very High: Decision Tree Model and ROC

Curve
Figure 7.5(a) presents the decision tree resulting from the best configuration of the C4.5 algorithm. As for the prediction of the variable
Relationship High, the first criteria of the tree is based on the metric
Frequency Quarter: provider employees who have established a very
high relationship with customer employees interacted with them in
at least three different quarters over the past year (Frequency Quarter >
50%). This condition is, however, not sufficient for the provider employee to consider having a very high quality relationship with the
customer employee: it is also necessary that either some face-to-face
interaction happened in the past three months (Mode 3M > 0) or that
a fairly high volume of interaction occurred with the customer before
the last year (Volume Weighted More1Y > 9.936).
These results lead to the following managerial implications: in order to develop very good relationships with customer employees,
the provider should try to develop long term projects with the customer, in which provider employees have to opportunity to interact
and especially meet in person with customer employees in at least
three of the four quarters of the year. This confirms that a transactional approach in which the provider employee meets the customer

242

7. CI Analytics Validation

employee only once or twice does not allow a development of qualitative relationships.
In addition, it is possible to draw the following conclusions from
the comparison of the results of the predictions of acquired knowledge
and established relationships obtained in sections 7.1.2 and 7.1.3. First,
the study reveals that a provider employee having a good knowledge of the customer employee has not automatically established a
good relationship with this customer employee. Reciprocally, having
established a qualitative relationship does not imply having a good
knowledge of the customer employee. Thus, this analysis confirms
the relevance of distinguishing acquired knowledge and established relationships at the individual level. Secondly, this analysis demonstrates
that acquiring knowledge of a customer employee requires a different pattern of interaction than to establish a qualitative relationship
with this employee. While the decision trees created to predict the
variable Knowledge High and Knowledge Very High emphasize the need
for frequent and intensive interactions in order to acquire customer
knowledge, the decision trees created to predict the variable Relationship High and Relationship Very High use the metric Frequency Quarter
in their first criteria, thereby highlighting the necessity for the provider employee to meet the customer in multiple quarters of the year
in order to establish qualitative relationships.
The next section of this chapter develops the results of the calibration
of the customer intimacy metrics to assess acquired knowledge and
established relationships at the organizational level.

7.2. Acquired Customer Intimacy at the
Organizational Level
While section 7.1 presents the results of the customer intimacy metrics calibration in order to predict the acquired customer intimacy
components at the individual level, this section details the calibration
to predict the acquired customer intimacy at the organizational level:
the objective is to assess to which extent a provider employee has
acquired some knowledge of, and established relationships with, a

7.2. Acquired Customer Intimacy at the Organizational Level

243

customer organization. Following the CI Analytics methodology presented in section 5.1.1 and the knowledge discovery in data mining
process illustrated in figure 3.2, the first part of this section focuses on
the data collection task. Then, the second and third parts present the
calibration results for the prediction of acquired knowledge and established relationships at the organizational level. Since these activities
have already been thoroughly described in section 7.1, this section
focuses on the main outcomes of the calibration and refers for details
to paragraphs in section 7.1.

7.2.1. Data Collection
The data collection tasks corresponds to the steps 4 and 5 of the
CI analytics methodology, which are the actual metric calculation
and the empirical assessment of the customer intimacy components.
These tasks are described in the two parts of this section.
7.2.1.1. Calculation of the Customer Intimacy Metrics
As explained in section 5.2.3, eight metrics have been designed upon
the concept of customer interaction time in order to assess customer intimacy at the organizational level. These metrics are volume, weighted
volume, intensity, weighted intensity, frequency, duration, number of episodes and mode of interaction. In addition, three network centrality metrics complement this list: the degree centrality, the normalized degree
centrality, and the normalized closeness centrality. In order to perform
the actual calculation, different parameters have to be determined.
The four parameter configurations determined for the calculation of
the metrics at the individual level and presented in section 7.1.1.1 are
reused for the calculation of the metrics at the organizational level.
The centrality based customer intimacy metrics are derived from the
graph representation of the customer intimacy metrics at the individual level: in order to calculate the centrality metrics, first the customer intimacy graph is created with the chosen customer intimacy
metric as a weighting function. Then, the centrality metrics values
are determined. In this scenario, the degree centrality, which reflects

244

7. CI Analytics Validation

the number of contacts of a provider employee in the customer organization is calculated upon the Volume 3M and Volume 12M graph
representations in order to determine the number of contacts over the
past 3 months and over the past year. The normalized degree centrality
is calculated upon the Volume 12M graph representation. Finally, the
normalized closeness centrality is calculated upon the Volume 12M and
Volume Weighted 12M graph representations.
Table 7.10 summarizes the 24 created metrics for the assessment of
the customer intimacy components. Similarly to the calculation of
the customer intimacy metrics at the individual level, the customer
intimacy metrics at the organizational level have been calculated for
all couples { p, o } where p represents a CAS employee, o represents
one of the 14 customers of CAS, and data shows that some interactions occurred between p and some employees of o in the past.
In order to perform the calculation, the prototypical application CI
Graph which is described in appendix E.3 has been used. This calculation resulted in a data set consisting of 398 records. Each record
contains a reference to a provider employee, a reference to a customer
organization, and the values of the 24 customer intimacy metrics.
Table 7.10.: Model Configurations and Metrics to Assess Acquired Cus-

tomer Intimacy at the Organizational Level
Configuration

A

B

C

D

Time Period T

3 Months

12 Months

12
Months

Over One
Year

Segment Size d

1 Month

1 Month

3 Months

N/A

Email CIT Value
demail

10 minutes

10 minutes

10
minutes

10
minutes

Letter CIT Value
dletter

10 minutes

10 minutes

10
minutes

10
minutes

Interaction Duration
Threshold ∆

1 Month

1 Month

1 Month

N/A

7.2. Acquired Customer Intimacy at the Organizational Level

245

Model Configurations and Metrics to Assess Acquired Customer Intimacy at the Organizational Level (Continued)
Metrics

A

B

C

D

Interaction Quantity
Threshold b

0

0

0

N/A

weighted Interaction
Quantity Threshold
wb

0

0

0

N/A

Volume

Volume 3M

Volume 12M

Volume
More1Y

weighted Volume

Volume
Weighted
3M

Volume
Weighted 12M

Volume
Weighted
More1Y

Intensity

Intensity
3M

Intensity 12M

weighted Intensity

Intensity
Weighted
3M

Intensity
Weighted 12M

Frequency

Frequency
3M

Frequency
12M

Duration

Duration
3M

Duration 12M

Number of Episodes

Number of
Episodes
3M

Number of
Episodes 12M

Mode of Interaction

Mode 3M

Mode 12M

Degree Centrality

Number of
Contacts
3M (based
on Volume
3M)

Number of
Contacts 12M
(based on
Volume 12M)

Metrics

Frequency
Quarter

246

7. CI Analytics Validation

Model Configurations and Metrics to Assess Acquired Customer Intimacy at the Organizational Level (Continued)
Metrics

Normalized Degree
Centrality

Normalized
Closeness Centrality

A

B

C

D

Degree
Centrality
12M (based on
Volume 12M)
Closeness
Centrality
12M (based on
Volume 12M)
Closeness
Centrality
Weighted 12M
(based on
Volume
Weighted 12M)

7.2.1.2. Empirical Assessment of the Customer Intimacy
Components
The empirical assessment of the customer intimacy components refers
to the step 5 of the CI Analytics methodology. At the organizational
level, the provider employees are asked to assess with a 7-point
Likert-type scale their knowledge of, and relationships with different
customer organizations with the following 6 items:
• Acquired knowledge of customer organizations
– Item 1.1: “My knowledge of [CompanyName]’s needs is
thorough.”
– Item 1.2: “I learned a lot about [CompanyName]’s preferences in the period I worked with it.”
– Item 1.3: “I know the customer [CompanyName] very
well.”

7.2. Acquired Customer Intimacy at the Organizational Level

247

• Established relationships with customer organizations
– Item 1.4: “As an employee, I have a high-quality relationship with [CompanyName].”
– Item 1.5: “As an employee, I have a very collaborative
relationship with [CompanyName].”
– Item 1.6: “I am satisfied with the relationship I have with
[CompanyName].”
Further details on the item selection is presented in section 4.3. The
use of Likert-type scales is motivated in section 3.1.3, and an illustrative questionnaire is presented in appendix A figure A.2.
CAS suggested 43 employees to participate to the empirical estimation, with the constraint that each employee performs a maximum
of three assessments at the organizational level in order to limit the
time investment. 127 records in the data set containing the calculated
customer intimacy metrics at the organizational level correspond to
these 43 CAS employees. These records are selected out of the 398
available records in order prepare the 43 questionnaires. The actual
survey was performed between October and November 2010. 25 out
of the 43 surveyed employees returned their questionnaire resulting
in 77 empirical assessments. As a result, the final data set to perform
the calibration of the metrics in order to assess the customer intimacy
components at the organizational level consists of 77 records. Each
record contains a reference to a CAS employee, a reference to one of
the 14 CAS customers, the 24 calculated customer intimacy metrics,
and the values of the six empirically assessed Likert items.

7.2.2. Calibration: Acquired Knowledge
This section presents the results of the calibration of the customer intimacy metrics in order to assess the customer intimacy component
acquired knowledge. This corresponds to the steps 6 and 7 of the CI
Analytics methodology. After the preprocessing and transformation
tasks are explained in the first two parts, the creation of machine
learning models and their validation are explained in the last two

248

7. CI Analytics Validation

parts of this section. Since these activities have already been thoroughly described in section 7.1.2, this section focuses on the main
outcomes of the calibration.
7.2.2.1. Preprocessing
The preprocessing activity consists of three main tasks:
• Anonymize the Data Set
The references to the respondents and to the customer organizations are removed from each record in the data set for the
reasons outlined in section 7.1.2.1.
• Manage Missing Values
As explained in section 7.1.2.1, there are two types of missing
values: first, the interactions which are not recorded in the customer information system, and which may influence the calculation of the customer intimacy metrics. Similarly to the calibrations at the individual level, no action is performed in order to
manage this type of missing values. Secondly, missing values
refer to the Likert items which were not assessed by the respondents in the scope of the empirical evaluation of the customer
intimacy components. The items 1.1, 1.2, and 1.3 were used in
order to assess the component acquired knowledge. The data set
contains only three missing values: the item 1.2 has not been
evaluated in three records. Following the method “Imputation
by Using Replacement Values” explained in section 7.1.2.1, the
value of the item 1.2 in these three records is calculated as the
average of the values of the items 1.1 and 1.3.
• Manage Outliers
The outliers are kept unchanged in the data set for the three
reasons explained in section 7.1.2.1.
7.2.2.2. Data Transformation
The objective of data transformation is to determine the target prediction values which are used for the calibration of the customer
intimacy metrics. Similarly to the other calibrations presented in this
thesis, the data transformation activity consists of two tasks:

7.2. Acquired Customer Intimacy at the Organizational Level

249

• Creation of the Summated Scale Knowledge
The summated scale Knowledge is created as the mean of V(Item
1.1), V(Item 1.2), and V(Item 1.3) which are the empirically assessed values of the previously defined Likert items 1.1, 1.2,
and 1.3. This scale is conceptually valid as these items have
already been used to assess knowledge in past literature, and
reliable as its Crombach’s alpha value is equal to 0.911 as illustrated in appendix D figure D.1
Knowledge =

V(Item 1.1) + V(Item 1.2) + V(Item 1.3)
3

(7.3)

• Knowledge Scale Binarization
The scale Knowledge consists of 19 ordinal classes that range
from the value 1 to the value 7 by increments of 0.33: {1, 1.33, ...,
6.66, 7}. Similarly to the data transformation applied at the individual level, the binarization method is applied in order to
convert this 19-class classification task into two 2-class classification tasks with the creation of two binary variables:
– Knowledge High: At the individual level, it is considered
that a record belongs to the class Knowledge High if the
value of the variable Knowledge is equal or above the median value of 4.5. At the organizational level, the variable
Knowledge can take the values 4, 4.33 and 4.66 but not 4.5.
Thus, the limit to distinguish the records of class Knowledge High at the organizational level is set to 4.66. The
variable Knowledge High is set to 1 if the value of the variable Knowledge is equal or above 4.66 and to 0 otherwise.
Within the calibration data set, 27 out of the 77 records belong to the class Knowledge High. This represents 35.1% of
the data set of 77 records.


Knowledge High =

1
0

if Knowledge ≥ 4.66
otherwise

250

7. CI Analytics Validation

– Knowledge Very High: Similarly to the Knowledge Very High
index at the individual level, a provider employee is considered as having a very high knowledge of the customer
organization if its average assessment of the items 1.1, 1.2,
and 1.3 is equal or above 6. Thus, the variable Knowledge
Very High is set to 1 if Knowledge is equal or above 6 and
to 0 otherwise. The number of records of class Knowledge
Very High in the data set is equal to 13. This represents
16.9% of the data set.


Knowledge Very High =

1
0

if Knowledge ≥ 6
otherwise

Table 7.11.: Proportions of Records of Class Knowledge High and
Knowledge Very High
Quantity of
Records with
Value 1

Quantity of
Records with
Value 0

Total
Quantity of
Records

Knowledge
High

27 (35.1%)

50 (64.9%)

77 (100%)

Knowledge
Very High

13 (16.9%)

64 (83.1%)

77 (100%)

The next parts of this section present the calibration results for the
prediction of the values of the variables Knowledge High and Knowledge Very High.
7.2.2.3. Knowledge High Calibration and Validation
In order to perform the calibration of the customer intimacy metrics, the four machine learning algorithms which are presented in
section 3.2.3 have been trained and tested with the 10 times 10-fold
crossvalidation method. These algorithms are the decision tree C4.5,

7.2. Acquired Customer Intimacy at the Organizational Level

251

the k-nearest neighbor algorithm, the support vector machine algorithm, and the multilayer perceptron with backpropagation neural
network. The following performance indicators are used to determine the calibration performance: precision, recall, success rate, Fmeasure, and Kappa statistic. These indicators are described in section 3.2.3. In order to facilitate the interpretation of these performance indicators values, the interpretation intervals good, fair, and
poor defined in section 7.1.2.3 are used. Table 7.12 summarizes the
ranges of these intervals. The actual values of the performance indicators are also detailed for all calibration results developed in this
section in order to ensure the completeness of this thesis.
Table 7.12.: Proposed Interpretation of the Performance Indicators
Interpretation
Good (%)

Fair (%)

Poor (%)

Precision

[80 − 100] [60 − 80[

[0 − 60[

Recall

[70 − 100] [50 − 70[

[0 − 50[

Success Rate

[75 − 100] [60 − 75[

[0 − 60[

F-Measure

[75 − 100] [55 − 75]

[0 − 55[

Kappa Statistic

[50 − 100] [40 − 50[

[0 − 40[

In order to identify the best configurations, each algorithm has been
trained and tested multiple times with different parameters. Table 7.13 presents the best results achieved with each of these algorithms and table D.1 in Appendix D provides further details on these
configurations. It can be observed in table 7.13 that the decision
tree C4.5, the k-nearest neighbor, and the support vector machine algorithm perform significantly better than the multilayer perceptron
neural network, even though they only achieve fair precision values
ranging from 73.0% to 75.0%. The support vector machine is clearly
better than the decision tree C4.5 and the k-nearest neighbor algorithm as its recall values is good with a value of 81.0%, while the
decision tree C4.5 and k-nearest neighbor algorithm only achieve re-

252

7. CI Analytics Validation

call values of 65.0% and 51.0%. The support vector machine also
achieves a good success rate of 82.1%, a good F-measure value of
75.0% and obtains a good Kappa statistic value of 61.0%.
Table 7.13.: Knowledge High: Performance Indicator Results (g=good;
f=fair; p=poor)
Model

Precision Recall

Success Rate F-measure

Kappa

(%)

(%)

(%)

(%)

(%)

C4.5

73.0 (f)

65.0 (f)

79.6 (g)

66.0 (f)

53.0 (g)

k-NN

73.0 (f)

51.0 (f)

78.9 (g)

58.0 (f)

47.0 (f)

SVM

75.0 (f)

81.0 (g) 82.1 (g)

75.0 (g)

61.0 (g)

NNBP

50.0 (p)

54.0 (f)

48.0 (p)

29.0 (p)

68.3 (f)

Figure 7.6(b) presents the ROC curve obtained with the best configuration of the decision tree C4.5. This figure indicates that this
algorithm performs fairly well in order to retrieve the first 60% of
the records of class Knowledge High as the corresponding false positive rate is below 20%. This performance, however, decreases when
the objective is to retrieve the remaining 40% of the records of class
Knowledge High.
The decision tree created with the best configuraton of the C4.5 algorithm is depicted in figure 7.6(a). This tree contains two criteria.
First, the tree verifies whether the value of the metric Volume Weighted 3M, which indicates the interaction quantity over the past three
months, is above 1.48 hours. If this is the case, then the decision tree
predicts that the record belongs to the class Knowledge High. Otherwise, the decision tree considers the value of the metric Number
of Episodes 12M. If the last 12 months contain at least three episodes,
then the variable Knowledge High is set to the value 1. From a managerial perspective, these results indicate that for a provider employee
to obtain some knowledge of a customer organization, the key aspect is that he spends some time working with this organization
(Volume Weighted 3M > 1.48). This confirms the results proposed in

7.2. Acquired Customer Intimacy at the Organizational Level

253

past literature and outlining that interaction quantity is positively associated with customer knowledge (Noorderhaven & Harzing, 2009,
p.2).
100
90

> 1.48

≤ 1.48

Number of Episodes
12M
≤2

True (17.0/2.0)
>2

80

True Positive (%)

Volume Weighted
3M

70
60
50

Area under ROC: 0.68
40
30

20
10

False (55.0/8.0)

True (5.0/1.0)

0
0

20

40

60

80

100

False Positive (%)

(a) Decision Tree Representation

(b) ROC Curve
33 Organization Level Knowledge High

Figure 7.6.: Knowledge High: Decision Tree Model and ROC Curve
7.2.2.4. Knowledge Very High Calibration and Validation
32 Organization
Level Knowledge
High results achieved with the four machine
Table
7.14 presents
the best
learning algorithms in order to predict the value of the variable
Knowledge Very High. Further details on the corresponding configurations are available in Appendix D table D.2. All algorithms achieve
a high success rate above 80.0%. However, none of the algorithms
obtains good precision and recall values. This indicates that the machine learning models are capable of predicting the value of the variable Knowledge Very High when this value is equal to 0, but not when
this value is equal to 1. The decision tree C4.5 and the k-nearest
neighbor algorithm obtain the highest precision with values of 41.0%
and 42.0%. These values remain too low as over half of the records
predicted as being of class Knowledge Very High are incorrectly classified. The decision tree achieves the best recall value, but this indicator remains too low with the value of 45.0%. Its Kappa statistic is
also low with a value of 35.0%.

Figure 7.7(b) illustrates the ROC curve obtained with the best configuration of the decision tree C4.5 algorithm. Importantly, even though

254

7. CI Analytics Validation

Table 7.14.: Knowledge Very High: Performance Indicator Results
(g=good; f=fair; p=poor)
Model

Precision

Recall

Success Rate F-measure

Kappa

(%)

(%)

(%)

(%)

(%)

C4.5

41.0 (p)

45.0 (p)

84.6 (g)

40.0 (p)

35.0 (p)

k-NN

42.0 (p)

39.0 (p)

87.6 (g)

38.0 (p)

36.0 (p)

SVM

32.0 (p)

35.0 (p)

80.1 (g)

32.0 (p)

23.0 (p)

NNBP

14.0 (p)

19.0 (p)

80.1 (g)

14.0 (p)

10.0 (p)

this curve confirms the poor performance of the prediction of the
variable Knowledge Very High, it also indicates that the model is effective in order to retrieve the first 40.0% of the records of class Knowledge Very High since the corresponding false positive percentage is
equal to 10.0%. Thus, this algorithm can be used if the objective is to
identify a few number of employees who have acquired a very high
knowledge of a customer organization.
Volume Weighted
3M
100

> 5.427

≤ 5.427

90

True (5.0)
> 6.8%

≤ 6.8%

False (66.0/4.0)

Volume More1Y
≤ 4.8

> 4.8

80

True Positive (%)

Mode of Interaction
3M

70
60
50

Area under ROC: 0.65
40
30

20
10

False (2.0)

True (4.0)

0
0

20

40

60

80

100

False Positive (%)

(a) Decision Tree Representation

(b) ROC Curve
35 Organization Level Knowledge Very High

Figure 7.7.: Knowledge Very High: Decision Tree Model and ROC
Curve
34 Organization Level Knowledge Very High

As illustrated in figure 7.7(a), volume and mode of interaction are the
two main interaction characteristics used by the decision tree in order

7.2. Acquired Customer Intimacy at the Organizational Level

255

to determine whether a record belongs to the class Knowledge Very
High. The tree should be interpreted with caution as it did not obtain
a good precision value. The first criteria of the tree uses the metric
Volume Weighted 3M, indicating thereby that the interaction quantity
within the last three month is important for a provider employee to
obtain a very good knowledge of a customer organization. If the
value of the metric Volume Weighted 3M is above 5.427 hours, then
the provider employee has a very good knowledge of the customer
organization. The second criteria of the tree is based on the metric
Mode of Interaction 3M. This criteria indicates that if less than 6.68%
of the interaction in the past three months occurred via face-to-face
meetings (Mode of Interaction 3M ≤ 6.8%), then the variable Knowledge Very High is set to 0, and the provider employee does not have a
very high knowledge of the customer organization. The third criteria
of the decision tree considers the metric Volume More1Y. If the previous condition on the mode of interaction is fulfilled, the provider employee is predicted as having a very high knowledge of the customer
organization if he already interacted with the customer organization
for more than 4.8 hours before the past year (Volume More1Y ≥ 4.8).
From a management perspective, these results confirm the calibration results obtained for the prediction of the variable Knowledge
High: if an organization wants to acquire some very good knowledge of its customers, it has to ensure that its employees have a high
volume of interaction with the customer employees. In addition,
these results show that a certain amount of face-to-face interaction is
necessary for obtaining this knowledge.

7.2.3. Calibration: Established Relationships
This section describes the results of the calibration of the customer
intimacy metrics in order to assess the customer intimacy component
established relationships at the organizational level. While the first and
second part of this section summarize the required preprocessing
and data transformation tasks, the third and fourth parts present the
actual results and their interpretation.

256

7. CI Analytics Validation

7.2.3.1. Preprocessing
Similarly to the preprocessing activity performed to assess the component acquired knowledge and presented in section 7.2.2.1, this preprocessing activity consists of three tasks:
• Anonymize the Data Set
References to the provider employee and customer organization in each record are removed since they are not required to
perform the analysis.
• Manage Missing Values
As for the other preprocessing tasks presented in this chapter, no corrective action is performed in order to manage the
missing interaction data in the provider’s information system.
The Likert items 1.4, 1.5 and 1.6 presented in section 7.2.1.2
have been empirically assessed in order determine the value
of the customer intimacy component established relationships at
the organizational level. Four out of the 77 records of the calibration data set do not contain an assessment of any of these
three items and, thus, are removed from the data set. All other
records contain the empirical assessment of all three items. The
dataset used to perform the calibration of the customer intimacy metrics to determine the value of the component established relationships at the organizational level therefore consists
of 73 records.
• Manage Outliers
Similarly to the calibrations presented in the previous sections,
the outliers are kept unchanged in the data set for the three
reasons explained in section 7.1.2.1.
7.2.3.2. Data Transformation
The data transformation activity relates to the transformation of the
empirical data into variables used to calibrate the customer intimacy
metrics in order to assess the customer intimacy component established relationships. This data transformation relates to the creation

7.2. Acquired Customer Intimacy at the Organizational Level

257

of the summated scale Relationship and its binarization with the creation of the variables Relationship High and Relationship Very High:
• Creation of the Summated Scale Relationship
With V(Item 1.4), V(Item 1.5), and V(Item 1.6) representing the
empirically assessed values of the items 1.4, 1.5, and 1.6, the
summated scale Relationship is calculated as the mean of these
three values. The conceptual validity of this scale is ensured
as the items were all already used in past literature in order
to assess relationship quality. This scale is also reliable as its
Crombach’s alpha value is equal to 0.891 as illustrated in appendix D figure D.1.
Relationship =

V(Item 1.4) + V(Item 1.5) + V(Item 1.6)
3

(7.4)

• Relationship Scale Binarization
The scale Relationship consists of 19 ordinal classes ranging from
the value 1 to the value 7 by increment of 0.33. The binarization process is performed in order to transform this 19-class
classification task into two 2-class classification tasks:
– Relationship High: The variable Relationship High is created
in order to identify the provider employees which have
established a high relationship with a customer organization. Similarly to the binary variable Knowledge High, the
variable Relationship High is set to 1 if Relationship is equal
or above the value 4.66 and to 0 otherwise. As presented in
table 7.15, the variable Relationship High is set to 1 in 35 out
of the 73 records of the calibration dataset, representing a
proportion of 45.5% of the calibration data set.

1
if Relationship ≥ 4.66
Relationship High =
0
otherwise
– Relationship Very High: The variable Relationship Very High
distinguishes the records indicating that a provider employee has established a very high relationship with a cus-

258

7. CI Analytics Validation

tomer organization from the other records in the data set.
It is set to 1 if the variable Relationship is equal or above 6
and to 0 otherwise. 16 out of the 73 records in the dataset
fulfil this condition and belong to the class Relationship
Very High. This represents a proportion of 21.9% of the
calibration data set, as illustrated in table 7.15.

Relationship Very High =

1
0

if Relationship ≥ 6
otherwise

Table 7.15.: Proportions of Relationship High and Relationship Very
High Records
Quantity of
Records with
Value 1

Quantity of
Records with
Value 0

Total
Quantity of
Records

Relationship
High

35 (44.5%)

38 (55.5%)

73 (100%)

Relationship
Very High

16 (21.9%)

57 (78.1%)

73 (100%)

7.2.3.3. Relationship High Calibration and Validation
Table 7.16 presents the best results obtained with the four previously
introduced algorithms. Further information on these configurations
is available in appendix D table D.3. Even though multiple configurations were tested, the three algorithms decision tree C4.5, k-nearest
neighbor, and the multilayer perceptron with backpropagation neural network all achieved a poor precision and a fair recall. They also
do not obtain a good success rate and the Kappa statistic is marginal
as it ranges between 11.0% and 17.0%. However, the support vector
machine algorithm obtains fair to good results to predict the variable
Relationship High. It obtains a fair precision of 64.0% and a good recall value of 74.0%. Its success rate is fair with a value of 68.8% but
its Kappa statistic value remains only poor with a value of 33.0%.

7.2. Acquired Customer Intimacy at the Organizational Level

Table 7.16.: Relationship High:
Performance
(g=good; f=fair; p=poor)

259

Indicator

Results

Model

Precision (%) Recall (%) Success Rate (%) F-measure (%) Kappa (%)

C4.5

55.0 (p)

55.0 (f)

56.0 (p)

52.0 (p)

11.0 (p)

k-NN

53.0 (p)

61.0 (f)

56.5 (p)

55.0 (f)

13.0 (p)

SVM

64.0 (f)

74.0 (g)

66.8 (f)

66.0 (f)

33.0 (p)

NNBP

54.0 (p)

67.0 (f)

58.6 (f)

58.0 (f)

17.0 (p)

The ROC curve of the multilayer perceptron with backpropagation
neural network is presented in figure 7.8(b). This curve confirms the
poor performance of the algorithm as the true positive rate is never
significantly higher than the false positive rate.
100
90
80

True Positive (%)

Closeness Centrality
12M
> 0.768

≤ 0.768

False (25.0/5.0)

Frequency_12m
≤ 8.33%

70
60
50

Area under ROC: 0.56
40
30
20

> 8.33%

10

False (11.0/4.0)

True (37.0/11.0)

0
0

20

40

60

80

100

False Positive (%)

(a) Decision Tree Representation

(b) ROC Curve
37 Organization Level Relationship High

Figure 7.8.: Relationship High: Decision Tree Model and Multilayer
Perceptron ROC Curve
Figure 7.8(a) presents the decision tree model created with the decision tree C4.5 algorithm. This model should be interpreted cautiously, as the algorithm did not achieve a particularly good perfor36 Organization Level Relationship High
mance. Importantly, the first criteria of the tree is based on the metric
Closeness Centrality 12M. If the value of this metric is below 0.768 then
Relationship High is set to 0. Otherwise, the decision tree considers
in its second criteria the metric Frequency 12M. If the provider em-

260

7. CI Analytics Validation

ployee had interaction with employees of the customer organization
in at least two different months (Frequency 12M > 8.33%) then the
variable Relationship High is set to 1. These results confirm the relevance of complementing the customer interaction time based metrics
with network centrality based metrics: the topology of the social network formed by the provider and customer employees influences the
perception of having established a qualitative relationship from the
provider employee’s perspective.
7.2.3.4. Relationship Very High Calibration and Validation
The four machine algorithms have been trained and tested in multiple configurations in order to predict the value of the variable Relationship Very High at the organizational level. The best results are presented in table 7.17 and the corresponding configurations of the algorithms are detailed in appendix D table D.4. All algorithms achieved
a good success rate above 75.0%. None of them, however, achieved
good precision and recall values. The best model is obtained with
the decision tree C4.5 algorithm. This model has a precision of 48.0%
and a recall value of 49.0%. Different reasons can explain the poor
performance of this calibration. First, the current metrics are not
suited for the prediction of the variable Relationship Very High and
the model should be complemented with further metrics in order
to perform the calibration. Second, the considered machine learning algorithms are not suited and other algorithms should be trained
and tested. Third, the items used to assess established relationships at
the organizational level may have been incorrectly interpreted by the
participants to the survey. This leads to a wrong assessment of this
component and prevents the calibration of the metrics to predict the
value of the variable Relationship Very High.
The ROC curve of the model created with the decision tree C4.5 algorithm is presented in figure 7.9(b). This curve confirms the low
performance of the algorithm. The decision tree created with the best
configuration of the C.5 algorithm is depicted in figure 7.9(a). This
model should be interpreted with caution as it achieved a poor performance. Regularity based metrics such as Frequency 12M and Number of Episodes 12M are not included in this tree, but Degree Centrality

7.2. Acquired Customer Intimacy at the Organizational Level

261

Table 7.17.: Relationship Very High: Performance Indicator Results
(g=good; f=fair; p=poor)
Model

Precision

Recall

Success Rate F-measure

(%)

(%)

(%)

(%)

(%)

C4.5

48.0 (p)

49.0 (p)

78.2 (g)

44.0 (p)

33.0 (p)

k-NN

41.0 (p)

29.0 (p)

79.6 (g)

33.0 (p)

24.0 (p)

SVM

32.0 (p)

35.0 (p)

80.1 (g)

32.0 (p)

23.0 (p)

NNBP

18.0 (p)

16.0 (p)

77.6 (g)

16.0 (p)

11.0 (p)

Volume Weighted
3M

100

> 5.427

≤ 5.427

90
80

Mode of Interaction
3M

True Positive (%)

True (5.0)
> 0.068

≤ 0.068

Degree Centrality
12M

False (66.0/7.0)

Kappa

≤ 0.089

70
60
50

Area under ROC: 0.67
40
30
20

> 0.089

10

True (3.0)

False (3.0/1.0)

0
0

20

40

60

80

100

False Positive (%)

(a) Decision Tree Representation

(b) ROC Curve
39 Organization Level Relationship Very High

Figure 7.9.: Relationship Very High: Decision Tree Model and ROC
Curve
38 Organization Level Relationship Very High

12M is one of the three considered metrics. The first and second criteria of this tree use the 3-month based variables Volume Weighted 3M
and Mode of Interaction 3M. Thus, the tree considers that the employees who interacted with the customer within the last three months
and who had some face to face interaction are those who established
very qualitative relationships with the customer organization. Importantly, this tree confirms the relevance of using network centrality
based metrics in order to assess the customer intimacy components
at the organizational level as the metric Degree Centrality 12M is used

262

7. CI Analytics Validation

by the tree in the third criteria.

7.3. Summary and Interpretation of the
Calibration Results
Sections 7.1 and 7.2 developed the results of the customer intimacy
metrics calibration for assessing the customer intimacy components
acquired knowledge and established relationships at the individual and
organizational levels. This section summarizes these results and further elaborates on their interpretation and managerial implications.

7.3.1. Results Summary
Table 7.18 details the best results achieved for each of the eight performed calibrations and confirms the effectiveness of the CI Analytics
methodology to estimate the values of the customer intimacy components:
• According to the interpretation interval specified in table 7.4,
three out of the eight calibrations show a good precision value
above 80.0% and three of them a fair precision value comprised
between 60.0% and 80.0%. Four calibrations achieve good recall
values above 70.0% and two of them a fair recall value inside
the 50.0% - 70.0% range.
• Six calibrations achieve a good success rate above 75.0%. The
remaining two calibrations obtain a fair success rate comprised
between 60.0% and 75.0%. With regard to the Kappa statistic
indicator, three calibrations present good values above 50.0%
for this indicator, and two calibrations a fair value in the 40.0%
- 50.0% range.
• Two calibrations which concern the prediction of the variables
Knowledge Very High and Relationship Very High at the organizational level lead to poor results. These calibration achieve precision and recall values comprised between 41.0% and 49.0%.
Different reasons may explain this phenomenon: the sample

7.3. Summary and Interpretation of the Calibration Results

263

Success Rate (%)

F-measure (%)

Kappa statistic (%)

Organization Level
Knowledge High
Knowledge Very
High
Relationship High
Relationship Very
High

Recall (%)

Individual Level
Knowledge High
Knowledge Very
High
Relationship High
Relationship Very
High

Precision (%)

Predicted Variable

Algorithm

Table 7.18.: Summary of the Calibration Results (g=good; f=fair;
p=poor)

NNBP

87.0 (g)

71.0 (g)

80.2 (g)

76.0 (g)

60.0 (g)

k-NN

83.0 (g)

57.0 (f)

83.5 (g)

64.0 (f)

55.0 (g)

k-NN

80.0 (g)

75.0 (g)

73.3 (f)

76.0 (g)

45.0 (f)

C4.5

75.0 (f)

52.0 (f)

81.1 (g)

58.0 (f)

48.0 (f)

SVM

75.0 (f)

81.0 (g)

82.1 (g)

75.0 (g)

61.0 (g)

C4.5

41.0 (p)

45.0 (p)

84.6 (g)

40.0 (p)

35.0 (p)

SVM

64.0 (f)

74.0 (g)

66.8 (f)

66.0 (f)

33.0 (p)

C4.5

48.0 (p)

49.0 (p)

78.2 (g)

44.0 (p)

33.0 (p)

size is too small for an effective training of the machine learning algorithms, the metrics chosen for the calibration are not
suited and new metrics should be defined, or the items used
for the empirical assessment were incorrectly interpreted by the
respondents.
• The results are overall better for the assessment at the individual level than at the organizational level: while the four calibrations at the individual level obtain a fair or good precision,
only two out of the four calibrations at the organizational level
achieved a fair or good precision.
• The predictions of the variables Knowledge High and Relationship

264

7. CI Analytics Validation

High are better than those of the variables Knowledge Very High
and Relationship Very High at both the individual and organizational levels. This may be explained by the higher number of
records of type “High” in the dataset.
• Each of the four considered machine learning algorithms achieved the best overall results for at least one of the eight performed calibrations. The decision tree C4.5 achieved the best results three times, followed by the k-nearest neighbor algorithm
and the support vector machine which obtained the best results
twice. Finally, the multilayer perceptron with backpropagation
neural network obtained the best results once, for predicting
the value of the variable Knowledge High at the individual level.
• In order to ensure an optimized usage of the machine learning
algorithms, each algorithm has been trained on average with 48
different configurations, as illustrated in appendix B table B.5.
Thus, an average of 193 tests has been conducted for each predicted variable and a total of 1545 tests for the overall analysis.
This aspect guarantees the completeness of the results obtained
in this thesis.
Additional findings can be drawn from the analysis of the decision
tree models presented in the previous sections. The number of occurrences of each metric in all decision trees is detailed in table 7.19.
In this table, the metrics are sorted according to their corresponding
interaction pattern as proposed in table 5.1. Even though this table
does not take into account the position of the different metrics in the
decision trees, the following aspects are significant:
• At the individual level, 13 out of the 19 calculated customer
intimacy metrics are used in the decision trees, confirming the
importance of these different metrics. Confirming past literature presented in section 5.2.2 on the impact of interaction regularity on knowledge and relationship, the regularity based
metrics such as Frequency and Number of Episodes are the most
important metrics as they occur seven times in the decision
trees. Interaction quantity is also a significant customer intimacy indicator as the corresponding metrics occur four times

7.3. Summary and Interpretation of the Calibration Results

265

Table 7.19.: Number of Occurrences of the Metrics in the Decision
Tree Models
Number of Occurrences
Customer Intimacy Metric
Interaction Regularity

Individual
Level
7

Organizational
Level
2

Total
9

Frequency Quarter

3

Frequency 12M

2

Number of Episodes 3M

1

Number of Episodes 12M

1

1

2

4

4

8

Volume More 1Y

1

1

2

Volume Weighted 12M

1

1

2

Volume Weighted 3M

1

2

3

Volume Weighted More1Y

1

Interaction Quantity

Mode of Interaction
Mode 3M
Intensity
Interaction Intensity 12M
Network Centrality

3
1

3
1

1

1

2

3

1

2

3

1

1

1

1
2

2

Closeness Centrality 12M

1

1

Degree Centrality 12M

1

1

10

23

Total

N/A

13

in the decision trees. Finally, the two other interaction patterns
Mode of Interaction and Interaction Intensity are also used by the
decision trees, thereby confirming their relevance for the assessment of the customer intimacy components.
• At the organizational level, the results should be interpreted
with caution since the decision tree C4.5 algorithm did not

266

7. CI Analytics Validation

perform as well as at the individual level. 10 out of the 24
calculated metrics are used in the decision trees. However, it
cannot be concluded that the remaining metrics are irrelevant
since they may have been used by the other three algorithms.
Contrary to the calibrations performed at the individual level,
interaction quantity is the most important interaction pattern
as the corresponding metrics occur four times in the decision
trees. Then, the two interaction patterns interaction regularity
and mode of interaction as well as the network centrality metrics
are equally represented with two occurrences in the decision
trees. The decision trees created at the organizational level,
however, do not use metrics based on the pattern interaction
intensity.

7.3.2. Results Interpretation
Multiple managerial implications can be drawn from the results developed in this chapter with regard to the acquisition of customer
knowledge and to the establishment of customer relationships. As
explained in chapter 5, the CI Analytics methodology is calibrated
to the specific interaction patterns of the provider, which is in the
context of this scenario the company CAS. Since the machine learning models created in this chapter are based on data provided by
CAS, the following managerial implications are valid for CAS. Their
validity for other providers should be evaluated in future research.
First, considering the acquisition of customer knowledge, this thesis
shows that it is possible to thoroughly assess the degree of knowledge that a provider employee has acquired on customer employees
upon the customer intimacy metrics. According to the results presented in sections 7.1.2.3 and 7.1.2.4, the provider should ensure that
its employees frequently and regularly interact with the customer in
order to foster the acquisition of this knowledge. The results based
on the metric Frequency 12M show that provider employees should
ideally meet the customer employees in at least four different months
within a year in order to obtain a good knowledge of the customer
employees. The use of the metric Frequency Quarter in the decision

7.3. Summary and Interpretation of the Calibration Results

267

trees shows that this knowledge is further developed if the interactions are distributed along the four quarters of the year. Moreover,
the results based on the metric Intensity 12M show that the acquisition of knowledge about customer employees is fostered by events
such as workshops or consulting projects in which provider and customer employees have the opportunity to interact for a longer duration. The provider should, therefore, try to organize such events
with his customers.
Focusing on the acquisition of customer knowledge at the organizational level, the results presented in sections 7.2.2.3 and 7.2.2.4 show
that it is possible to effectively identify the first 60% of the provider
employees who have acquired a good knowledge of a customer organization. The results based on the metric Volume Weighted 3M outline
that the provider should make sure that his employees have a significant amount of interaction with the customer every quarter in order
to acquire a good knowledge of the organization. In addition, the use
of the metric Mode of Interaction 3M in the decision trees confirms the
importance of having a certain level of face-to-face interaction with
the customer in order to obtain this knowledge. Aligned with past
literature (Ballantyne, 2004; Hakansson et al., 2009), these results confirm the importance of interactions for the development of customer
knowledge.
With regard to the relationships established between provider and
customer employees, the ROC curves depicted in sections 7.1.3.4 and
7.1.3.4 demonstrate that the customer intimacy metrics support in a
very effective way the identification of the first 50% of the provider
employees who have established good or very good relationships
with customers. The predominance of the metrics Frequency Quarter
and Number of Episodes outlines the importance of the regularity in
the interactions in order to support the development of qualitative
relationships between provider and customer employees. Provider
employees should interact with the customer employees in at least
two different quarters in one year to establish a good relationship
and in at least three different quarters in one year to establish a very
good relationship.

268

7. CI Analytics Validation

At the organizational level, the results presented in sections 7.2.3.3
show that the topology of the social network formed provider and
customer employees has an important influence of the quality of the
relationship established by the provider employee with the customer
organization and, thus, further research should be performed in that
direction. The results, however, do not allow to draw further conclusions on how to support the development of relationships between a
provider employee and a customer organization.
The next chapter will conclude this thesis. It will summarize the
contribution of this thesis, develop the extent to which the research
questions defined in chapter 1 have been answered, and outline directions for future research.

8. Conclusion
This thesis was motivated by three main factors. First, customer intimacy has become over the past decades a prominent type of business
strategy. It receives a growing interest from businesses undergoing
a servitization endeavor and trying to generate competitive advantages from customer related knowledge and customer relationships.
Second, a literature review developed in chapter 4 demonstrates the
lack of methods, tools, and techniques enabling the assessment of
customer intimacy. From the IT perspective, even though CRM systems aim to support the management of the relationship, they do not
provide operational means and metrics for actually measuring the
degree of customer intimacy established with different customers.
Finally, the increasing importance of business analytics and social
network analysis techniques in both practice and academia together
with the availability of large scale database on which such analyses
can be performed was the third motivating factor of this thesis. Contemporary businesses seek new solutions based on such methods
and techniques to derive some knowledge out of the vast amount of
gathered customer related data and to support their decision-making
processes.
Combining these three factors, the central argument of this thesis is
that, in a B2B context, the customer intimacy achieved by a provider
organization with its different customers and its business impact can

270

8. Conclusion

be assessed and monitored at multiple levels of granularity – the individual and organizational levels – using social network analysis
and business analytics techniques. This assessment and monitoring
is achieved by leveraging customer related data available in the information system of the provider and by calculating a set of customer
intimacy metrics from this data. This chapter will examine to which
extent this argument has actually been validated by this thesis. Section 8.1 will revisit the three research questions defined in chapter 1
and will summarize the contribution of this thesis. Section 8.2 will
subsequently elaborates on the managerial implications of this thesis. Finally, section 8.3 will address the limitations of this thesis and
suggest directions for future research.

8.1. Contribution
In chapter 1, three main research questions addressing the central
argument of this thesis have been defined. This section summarizes
the solution proposed by this thesis to answer these questions and,
thereby, outlines the contribution of this thesis.
Research Question 1 – How can the concept of customer intimacy
be broken down into multiple assessable customer intimacy components?
Customer intimacy is a complex type of business strategy which aims
at achieving sustainable competitive advantages by intensifying customer relationships and utilizing customer knowledge. In order to
thoroughly evaluate this strategy, it is necessary to determine the
provider’s investments in developing his customer intimacy strategy
as well as the effectiveness of this strategy with the different customers. The first research question of this thesis is therefore concerned with how customer intimacy can be broken down into multiple assessable customer intimacy components.
This question is addressed in chapter 4. Starting from the original
definition of customer intimacy proposed by Treacy & Wiersema
(1993, p.87) – “to tailor and shape products and services in order

8.1. Contribution

271

to fit an increasingly fine definition of the customer” – this chapter
exploits findings in customer intimacy related literature in order to
determine the actual customer intimacy components. This thesis establishes that customer intimacy can be decomposed into two parts,
namely the acquired and leveraged customer intimacy, and argues that
both parts are required for the provider to successfully become “customer intimate” with his customers.
Acquired customer intimacy refers to obtaining this “fine definition
of the customer” and consists of acquiring customer knowledge and
establishing customer relationships. Customer knowledge is foundational to the development of a customer intimacy strategy because
a thorough understanding of the customer is required to be able
to adapt the solution provided to the customer. Customer knowledge covers multiple aspects such as the customer needs, satisfaction, expectations, strategy, and future plans. Established customer
relationships is a second cornerstone of the acquired customer intimacy as customer intimacy is grounded in the domain of relationship marketing. Customer relationships are particularly significant
in the considered context of B2B markets because they are an antecedent to customer knowledge: customer relationships allow the
provider to understand his customers and, therefore, to improve his
value proposition accordingly.
Leveraged customer intimacy reflects the actual benefits, competitive
advantages, and means to improve the value proposition that the
provider achieves by leveraging the acquired customer intimacy. It
corresponds to the part “to tailor and shape products and services”
of the definition of customer intimacy. The analysis performed in this
thesis upon existing literature has led to the identification of six components pertaining to the leveraged customer intimacy. These components are customization, customer loyalty, proactiveness, crossselling, customer participation, and transaction costs reduction. Using customer knowledge and customer relationships, the provider
can customize his solution to the needs of the customer, increase
customer loyalty, be proactive and anticipate the customer’s expectations, increase revenues through cross-selling, improve his offering
by involving the customer in the creation process, or reduce transac-

272

8. Conclusion

tional costs. The provider thereby generates a competitive advantage
or improves his value proposition.
This first research question is, thus, answered by this breakdown
analysis which has led to the identification of two components for
the acquired customer intimacy and six components for the leveraged
customer intimacy.
Research Question 2 – Which metrics can be created upon customer
related data in order to infer the customer intimacy components?
The second research question of this thesis concerns the definition
of metrics allowing the assessment of the customer intimacy components upon customer related data at both the individual and organizational levels. The solution to this question is elaborated in
chapter 5.
The inference challenge developed in chapter 5 is a central issue of
the assessment of the two acquired customer intimacy components acquired knowledge of, and established relationships with, customers.
This challenge roots in the fact that no means is well recognized and
established for analytically evaluating customer knowledge and customer relationships. In past literature, these components are mostly
assessed in an empirical way. To circumvent this issue, this thesis
proposes the CI Analytics model which relies on marketing literature
and associates these two concepts to the four interaction characteristics quantity, intensity, regularity, and mode. Eight metrics are subsequently derived from these characteristics based on the concept of
customer interaction time to assess acquired customer knowledge and
established customer relationships. These metrics are volume, weighted volume, intensity, weighted intensity, frequency, duration, and number
of episodes. At the organizational level, three additional metrics leveraging the topology of the social network formed by the provider
and customer employees are defined. These metrics are the degree
centrality, the normalized degree centrality, and the normalized closeness
centrality.
In order to evaluate the leveraged customer intimacy components, this
thesis elaborates a set of eight metrics by investigating prior research

8.1. Contribution

273

and analyzing sources of data which are relevant for their assessment. These eight metrics are customization revenue ratio, customer
purchase frequency ratio, proactiveness ratio, cross-selling revenue ratio,
cross-selling diversity ratio, customer participation quantity, customer participation ratio, and transaction effectiveness ratio. The calculation of
these metrics occurs upon interaction, activity, and revenue records.
To validate the feasibility of the calculation of these customer intimacy metrics and to make them available to users, the software CI
Analytics which is detailed in chapter 6 has been conceived and implemented in the scope of this thesis. This software is built upon
business intelligence applications standards, storing the relevant interaction, activity, and revenue data in a data warehouse. The software CI Analytics supports in its current version the calculation of
the customer intimacy metrics upon the data contained in the application genesisWorld from CAS Software AG (CAS). Since the data
contained in the warehouse can be updated on a regular basis at a
user defined frequency, the software CI Analytics provides, in addition to the calculation of the customer intimacy metrics, the ability
to monitor the evolution of these metrics over time.
In order to answer the second research question, this thesis, thus,
establishes eight metrics to assess the acquired customer intimacy at
the individual level, 11 metrics to assess acquired customer intimacy
at the organizational level and eight metrics to assess the leveraged
customer intimacy components. Moreover, this thesis confirms the
feasibility of the assessment and monitoring of these metrics through
the realization of the software CI Analytics.
Research Question 3 – Which combination of metrics provides the
most accurate assessment of the customer intimacy components?
The third research question concerns the selection of the most relevant customer intimacy metrics and their combination in order to
effectively assess the customer intimacy components. This question
raises two issues which are the determination of the relevance of the
different metrics and the calibration of the metrics to fit the interaction
and activity patterns of each provider.

274

8. Conclusion

The CI Analytics methodology which is developed in chapter 5 is the
solution proposed by this thesis to these two issues. This methodology is based on the established knowledge discovery in database process
which outlines the required steps for analyzing data contained in
databases (Fayyad et al., 1996a). The CI Analytics methodology requires on the one hand to perform an empirical assessment of the
customer intimacy components for selected customers by means of
a survey with provider employees and on the other hand to calculate the customer intimacy metrics for the same customers. The
two resulting data sets are subsequently merged in order to perform
a supervised data-mining analysis. In this analysis, the calculated
metrics are the prediction variables and the results of the empirical assessment are transformed into the predicted variables. Several
machine learning algorithms are trained to predict the empirically
assessed values of the customer intimacy components upon the calculated customer intimacy metrics. The resulting models are finally
tested to ensure that they can be successfully applied to other data
from the same provider, and interpreted in order to understand the
most relevant metrics and to derive managerial implications.
The CI Analytics methodology has been validated in a real-case scenario with the company CAS. The results are detailed in chapter 7.
The components acquired customer knowledge and established customer relationships related to 14 different customers were assessed
by CAS employees and the corresponding customer intimacy metrics
were calculated with the software CI Graph outlined in appendix E.3.
Four algorithms have been trained to predict the empirically assessed
customer intimacy components upon the calculated customer intimacy metrics: the decision tree C4.5, the multilayer perceptron with
back propagation neural network, the k-nearest neighbor algorithm,
and the support vector machine algorithm. The results have been
evaluated using the 10-fold cross-validation technique with the performance indicators precision, recall, success rate, F-measure, and
Kappa statistic.
The results developed in chapter 7 show that eight calibrations of
the customer intimacy metrics have been performed to predict the
acquired customer intimacy components at the individual and orga-

8.2. Managerial Implications

275

nizational levels. Six calibrations achieve a good success rate. Six
calibrations achieve good or fair precision and recall values. Overall, the results of the calibration at the individual level are better
than those at the organizational level. The four machine learning algorithms performed differently but none of them was significantly
better than the others. The interaction metrics based on regularity
such as frequency and number of episodes are the most relevant ones
for assessing the customer intimacy components at the individual
level. At the organization level, the interaction quantity metrics such
as volume and weighted volume are the most significant ones.
This analysis confirms the effectiveness of the CI Analytics methodology for determining the best combination of metrics to assess the
acquired customer intimacy components. While the implementation
of the software CI Analytics proves the feasibility of calculating, monitoring, and representing the proposed customer intimacy metrics, the
quantitative results validate the central argument of this thesis and
demonstrate that it is possible to accurately assess customer intimacy
at multiple level of details in an analytical manner.
The next section of this chapter elaborates on the managerial implications of the results obtained in this thesis.

8.2. Managerial Implications
The results achieved in the course of this thesis may have in the
future significant managerial implications as they allow an organization pursuing a customer intimacy strategy to obtain new insights in
the actual development and implementation of this strategy with its
customers.
First, the software CI Analytics conceived in this thesis and described
in chapter 6 can be used by a provider in order to assess the degree of customer intimacy established with its different customers at
different levels of details and, thus, to support the future investments
and business decisions. As illustrated in figure 6.4, this software allows on one side to assess the investments performed by the provider

276

8. Conclusion

employees in order to acquire knowledge of, and establish relationships with, the customer, and on the other side, using the leveraged
customer intimacy indicators, to assess the business impact of this
knowledge and of these relationships. In a best-case scenario, as outlined in figure 4.1, a provider pursuing a customer intimacy strategy
should see in the CI Analytics dashboard high values with regard to
acquired knowledge and established relationships as well as high values with regard to the leveraged customer intimacy metrics, thereby
indicating that the provider effectively used his knowledge of, and
relationships with, customers in order to derive competitive advantages and to improve its value proposition. However, if the CI Analytics dashboard indicates high knowledge and relationships values
but low leveraged customer intimacy values, then the customer intimacy strategy is not effective as no or few competitive advantages are
derived from the acquired knowledge and established relationships.
In such cases, the provider should analyze whether the customer intimacy strategy is appropriate with the customer as some customers
are not responsive to a customer intimacy strategy and are not ready
to pay a premium for a tailored solution. The provider should also
analyze whether the customer intimacy strategy was correctly implemented with this specific customer. It is indeed possible that an appropriate solution was not suggested to the customer, leading to low
leveraged customer intimacy values. Since the metrics can be used
in order to determine the customers with whom the customer intimacy strategy was most effective, this approach, in addition, allows
a ranking and benchmarking of the different customers, thereby supporting the provider with regard to its future customer investments.
The second type of managerial implications relates to an improved
coordination of the customer facing activities of the provider employees and a better sharing of customer knowledge inside the provider
organization. By making the values of the customer intimacy metrics available inside the provider organization, for instance in the
form proposed by the CI Analytics dashboard, the provider employees can easily identify colleagues who have acquired knowledge
of, and established relationships with, the customer as well as those
who were in contact with specific customer employees within a spe-

8.2. Managerial Implications

277

cific time frame, such as the past week or the past month. Using this
information, the provider employees whose activities are related to
a specific customer can find each other, exchange their knowledge,
and coordinate their activities. For instance, if a provider employee
p1 has planned a meeting with a customer employee c and notices
in the CI Analytics dashboard that another provider employee p2 had
a conversation with c in the past week, p1 can contact p2 to obtain
the most recent information on c and use this knowledge when he
meets c, thereby optimizing the interaction flow with the customer
employee c.
Finally, the approach proposed by this thesis allows an organization to gain insights on how to best establish, maintain, and enhance customer relationships as well how to effectively acquire customer knowledge by optimizing the customer interactions and activities. The results presented in section 7.3.2 shows that the company
CAS whose data was used to apply the CI Analytics methodology
should focus on specific interaction patterns in order to acquire customer knowledge and establish customer relationships. For instance,
CAS employees willing to acquire a good knowledge of customer
employees should interact with them in at least four different months
within a year. Moreover, in order to obtain a very good knowledge,
they should organize events of longer durations. To establish qualitative relationships, a focus should be given to the regularity of the
interaction: the provider employees should interact in three different
quarters of the year with customer employees in order to establish
very good relationships. These conditions are naturally not sufficient for acquiring knowledge and establishing relationships. An
employee interacting in three different quarters does not always have
a very good relationship with customer employees, but the probability that he does are higher if he follows these interaction patterns.
The next section of this chapter outlines the limitations of this contribution and suggests some directions for future research.

278

8. Conclusion

8.3. Outlook on Future Research
This thesis demonstrates that customer intimacy can be assessed and
monitored at multiple levels of details in a B2B context using business analytics and social network analysis methods. It is also laying
the foundations for further research investigating customer intimacy,
relationships, and business performance in an analytical way. This
section develops the limitations of the current approach and elaborates on future paths of research which could be followed upon this
thesis. Seven main aspects have been identified.
• Use different data sources to calculate the customer intimacy
metrics
The software CI Analytics which has been conceived and implemented in the scope of this thesis is able to process data
contained in the application CAS genesisWorld. A key benefit of CAS genesisWorld is that the relevant customer interaction, activity, and revenue data is stored in one single
database with appropriate references to customers and customer employees. However, because the software CI Analytics
only focuses in its current version on data contained in this
database, the proposed CI Analytics methodology has not been
applied to, and tested with, other sources of data. Future research should, therefore, concentrate on the integration of new
sources of data in the proposed approach to assess customer intimacy and the next version of the software CI Analytics should
support the access and processing of data contained in additional data sources such as CRM software, groupware, and
project databases. This task is facilitated by the current architecture of the software CI Analytics which allows an easy
integration of different sources of data.
• Develop additional customer intimacy metrics to improve the
assessment of the customer intimacy components
As developed in chapter 7, most of the performed calibrations
to assess the customer intimacy components achieved good or
fair results. However, some of these calibrations did not obtain
acceptable results with regard to the five defined performance

8.3. Outlook on Future Research

279

indicators. For instance, the precision and recall values related
to the prediction of the variable Relationship Very High at the
organizational level only obtained poor results even though the
corresponding success rate is good. Future research should,
therefore, investigate the creation of new metrics to complement the existing ones and to improve the quality of the performed customer intimacy assessment. In particular, activity
based metrics focusing on the time spent by customer employees on customer projects should be developed. Such data may
easily be retrieved from project databases. In addition, different calibration parameters such as the time period T, the segment size d, or the interaction duration threshold ∆ have been
proposed by this thesis in order to configure the calculation
of the customer intimacy metrics. In this thesis, as detailed
in chapter 7, four different configurations of these parameters
have been considered at both the individual and organizational
levels. Future research should test additional configuration as
well as further investigate the impact of these parameters on the
accuracy of the metrics to assess the customer intimacy components.
• Perform longitudinal analysis and add complex event processing
The software CI Analytics provides the means to calculate the
customer intimacy metrics at regular time intervals, thereby enabling the monitoring of the proposed customer intimacy components. The validation performed in the scope of this thesis
and elaborated in chapter 7, however, only considers a specific
point in time in order to calculate the metrics. Future research
should consequently focus on a longitudinal analysis of the customer intimacy metrics and evaluate which knowledge can be
derived from this time driven analysis. This analysis would, for
instance, uncover correlations between the evolution of the interaction and activity based metrics and business results. Such
research could subsequently be combined with complex event
processing in order to identify specific patterns among interaction and activity events which impact business activities (Et-

280

8. Conclusion

zion & Niblett, 2010). For instance, a change of the interaction
regularity combined with a drop of the activity volumes could
indicate some issues with the customer which should be proactively managed by the provider.
• Investigate the correlation between the acquired and leveraged customer intimacy components and conceive a recommender system based on successful interaction and activity
patterns
This thesis establishes a model to decompose customer intimacy into multiple components and develops multiple metrics
enabling the assessment of these components upon interaction,
activity, and revenue data. However, the analysis of correlations among the different customer intimacy components was
out of the scope of this thesis. Future research focusing on these
correlations is an important research topic potentially having
significant managerial implications.
First, focusing on the acquired customer intimacy, an investigation of the correlation between the acquired knowledge of,
and the established relationships with, customers would provide an understanding of the influence of customer relationships on acquired customer knowledge. Second, focusing on
the causal relationship between the acquired and leveraged customer intimacy, this investigation would provide insights on
which degrees of customer knowledge and customer relationships are required in order to reach the benefits elaborated in
the leveraged customer intimacy components. This analysis
can be performed analytically rather than empirically using
the proposed customer intimacy metrics. It would, therefore,
provide a unique contribution by associating some specific interaction and activity patterns, such as the regularity or the
volume of interactions to critical business impact factors such
as cross-selling revenues, customer loyalty, or transaction costs
reduction.
These patterns could subsequently be implemented into a recommender systems which supports the determination of the

8.3. Outlook on Future Research

281

customer related activities of the provider. For instance, if
the analysis establishes that a specific frequency of interaction
has an impact on customer loyalty, the system could remind
the corresponding provider employees to contact the customer
employees at this frequency. If a specific incentive has been
identified as particularly successful for facilitating opportunity
closure and for reducing transaction costs, this incentive could
be suggested to other provider employees which are in similar
situations with their customers.
• Elaborate a recommender system for optimizing the team in
charge of a customer, for allocating provider employees to
customer projects, and for coordinating the activities of these
employees
The approach developed in this thesis provides the means to assess and monitor the degree of customer intimacy established
with different customers. It also supports the exchange of customer related knowledge through the visualization of the social
network formed by the provider and customer employees upon
their interactions and joint activities.
This approach could be further extended in future research by
conceiving a recommender system which suggests a set of provider employees which are most likely to fit with the customer
organization upon the customer intimacy metrics measured at
the individual and organizational levels. Considering a specific
customer, this recommender system could consider as inputs
the roles and positions of the provider and customer employees, the current values of the acquired and leveraged customer
intimacy metrics, and the objectives set by the provider for this
customer. In return, this recommender system could provide a
set of employees which have the adequate skills as well as the
appropriate relationships and customer knowledge in order to
effectively and successfully perform the customer project. This
system would therefore support the optimization of the teams
in charge of specific customers and the allocation of the provider employees to the different customer projects.

282

8. Conclusion

Moreover, since the customer intimacy metrics are monitored
and updated at frequent intervals, this recommender system
can easily gather details on the most recent interactions and
activities that occurred with the customer employees. This system could therefore use this information in order to make recommendations to the provider employees before they contact
customer employees, thereby supporting the coordination of
the customer-facing activities. For instance, if a provider employee recently worked with several customer employees, the
other provider employees should contact him prior to contacting this customer employees as he may have some valuable
information and knows the details of the communication with
the customer. This could be automatically supported by this
recommender system.
• Evaluate the legal aspects of the customer intimacy assessment
A critical aspect of the assessment and monitoring of the degree of customer intimacy resides in the use of personal interaction records such as emails or details on meetings. Under
German law, this data does not belong to the provider organization but to the provider and customer employees involved
in the corresponding interactions who for instance send and receive the emails. The provider organization is, thus, not directly
allowed to use this data in order to perform the customer intimacy assessment. This problem is solved in this thesis through
the exclusive use of data stored in the application CAS genesisWorld. Provider employees can freely decide for each interaction record whether they want it to be transfered to CAS
genesisWorld. If the record is transfered to CAS genesisWorld,
it is then considered as a business information and can be used
by the organization. However, in order to access data contained
in other sources of data such as email servers, a legal solution
should be found. Thus, further research should further investigate from a legal perspective how to enable the calculation and
utilization of the customer intimacy metrics in the provider organization.

8.3. Outlook on Future Research

283

• Extend the proposed model towards B2C and C2C businesses
The model proposed by this thesis focuses on B2B organizations and takes into account the specific constraints of B2B
businesses, such as the fact that users and purchasers of the
provided solutions are different individuals in the customer organization. However, considering the size of B2C markets and
the increasing importance of B2C and C2C services in mature
economies, future research should focus on the extension of
this approach towards B2C and C2C businesses and the development of B2C and C2C specific customer intimacy metrics. This approach could subsequently be integrated in Internet based social network applications such as LinkedIn, Facebook, or Xing.
Following an interdisciplinary approach, this thesis proposes a novel
means for the assessment and monitoring of customer intimacy, combining a strategy and marketing concept with business analytics, network analysis, and software engineering. The outlook on future research developed in section 8.2 demonstrates the significance of the
managerial implications of this approach and shows that this thesis
lays the foundation for a wide variety of new research topics and
for a new way to approach the assessment and implementation of
business strategies.

Bibliography
Abraham, Gerard A. 2006. Profit Power of Customer Intimacy: Deliver Top Line Revenue and Earnings Growth. http://www.leadervalues.com/Content/detail.asp?ContentDetailID=1125, accessed
on 20/11/2011.
Adler, Lee. 1966. Symbiotic marketing. Harvard Business Review,
45(2), 59–71.
Akura, M. Tolga, & Srinivasan, Kannan. 2005. Customer Intimacy
and Cross-Selling Strategy. Management Science, 51(6), 1007–1012.
Ali, Shawkat, & Smith, Kate A. 2006. On learning algorithm selection
for classification. Applied Soft Computing, 6(2), 119–138.
Alpaydin, Ethem. 2010. Introduction to Machine Learning (Adaptive
Computation and Machine Learning series). The MIT Press.
Anderson, Erin, & Weitz, Barton A. 1992. The Use of Pledges to
Build and Sustain Commitment in Distribution Channels. Journal
of Marketing Research, 24(February), 18–34.
Arndt, Johan. 1979. Towards a Concept of Domesticated Markets.
Journal of Marketing, 43(4), 69–75.
Asratian, Armen S., Denley, Tristan M. J., & Häggkvist, Roland. 1998.
Bipartite graphs and their applications. Cambridge University Press.
Ballantyne, David. 2004. Dialogue and its role in the development
of relationship specific knowledge. Journal of Business & Industrial
Marketing, 19(2), 114–123.

286

Bibliography

Barnes, James G. 1997. Closeness, strength, and satisfaction: Examining the nature of relationships between providers of financial
services and their retail customers. Psychology and Marketing, 14(8),
765–790.
Batt, Peter. 2004. Managing collaboration within networks and relationships. Industrial Marketing Management, 33(3), 169–174.
Bauer, Kent. 2004. The KPI Profiler: CRM Case Study. DM Review.
Beauchamp, Murray A. 1965. An improved index of centrality. Behavioral Science, 10(2), 161–163.
Bennett, Rebekah, & Bove, Liliana L. 2001. Identifying the Key Issues
for Measuring Loyalty. Australasian Journal of Market Research, 9,
27–44.
Bennett, Robert, & Robson, Paul. 1999. Intensity of Interaction in Supply of Business Advice and Client Impact: a Comparison of Consultancy,
Business Associations and Government Support Initiatives for SMEs.
Berger, Paul D., Bolton, Ruth N., Bowman, Douglas, Briggs, Elten,
Kumar, V., Parasuraman, A, & Creed, Terry. 2002. Marketing Actions and the Value of Customer Assets: A Framework for Customer Asset Management. Journal of Service Research, 5(1), 39–54.
Berry, Leonard L. 1983. Relationship Marketing. Pages 25–38 of: Berry,
Leonard T., Shostak, G Lynn, & Upah, Gregory D. (eds), Emerging
Perspectives on Services Marketing. Chicago: American Marketing
Association.
Berry, Leonard L. 1995. Relationship Marketing of Services - Growing
Interest, Emerging Perspectives. Journal of the Academy of Marketing
Science, 23(4), 236–245.
Berry, Michael J. A., & Linoff, Gordon S. 2004. Data Mining Techniques:
For Marketing, Sales, and Customer Relationship Management. 2 edn.
Wiley Computer Publishing.

Bibliography

287

Bettencourt, Lance A. 1997. Customer voluntary performance: Customers as partners in service delivery. Journal of Retailing, 73(3),
383–406.
Biemans, Wim G., Brencic, Maja, & Malshe, Avinash. 2010. Marketing
sales interface configurations in B2B firms. Industrial Marketing
Management, 39(2), 183–194.
Blois, Robert. 2008. CRM Spending up, but User Adoption Still
a Problem.
http://www.itbusinessedge.com/cm/community/
features/interviews/blog/crm-spending-up-but-user-adoptionstill-a-problem/?cs=22697, accessed on 20/11/2011.
Boles, James, Johnston, Wesley, & Gardner, Alston. 1999. The selection and organization of national accounts: a North American
perspective. Journal of Business & Industrial Marketing, 14(4), 264–
282.
Bolton, Ruth N. 2006. Foreword. Page ix of: Lusch, Robert F., &
Vargo, Stephen L. (eds), The Service-Dominant Logic of Marketing.
M.E.Sharpe.
Bove, Liliana L., & Johnson, Lester W. 2001. Customer relationships with service personnel: do we measure closeness, quality
or strength? Journal of Business Research, 54(3), 189–197.
Brandes, Ulrik, & Erlebach, Thomas. 2005a. Fundamentals. Chap. 2,
pages 7–15 of: Brandes, Ulrik, & Erlebach, Thomas (eds), Network
Analysis: Methodological Foundations. Springer.
Brandes, Ulrik, & Erlebach, Thomas. 2005b. Introduction. Chap. 1
of: Brandes, Ulrik, & Erlebach, Thomas (eds), Network Analysis:
Methodological Foundations. Springer.
Bruhn, Manfred, & Georgi, Dominik. 2006. Services marketing: managing the service value chain. 1 edn. Pearson Education.
Buechel, Berno, & Buskens, Vincent. 2008. The Dynamics of Closeness
and Betweenness.

288

Bibliography

Bueren, Adrian, Schierholz, Ragnar, Kolbe, Lutz, & Brenner, Walter.
2004. Customer knowledge management - improving performance
of customer relationship management with knowledge management. 37th Annual Hawaii International Conference on System Sciences,
2004. Proceedings of the, 00(C), 172–181.
Burton, Betsy, Genovese, Yvonne, Rayner, Nigel, Casonato, Regina,
Smith, Michael, Beyer, Mark A, Austin, Tom, Gassman, Bill, &
Sommer, Dan. 2011. Predicts 2011 : Pattern-Based Strategy Technologies and Business Practices Gain Momentum. Gartner Inc., 1–
15.
Cannon, Joseph P., & Narayandas, Narakesari. 2000. Relationship
Marketing and Key Account Management. Chap. 15, pages 407–
429 of: Sheth, Jagdish N., & Parvatiyar, Atul (eds), Handbook of
Relationship Marketing, 1 edn. Sage Publications, Inc.
Cermak, Dianne S.P., & File, Karen Maru. 1994. Customer participation in service specification and delivery. Journal of Applied Business
Research, 10(2), 90–100.
Chapman, Pete, Clinton, Julian, Kerber, Randy, Khabaza, Thomas,
Reinartz, Thomas, Shearer, Colin, & Wirth, Rüdiger. 2000. Step-bystep data mining guide. Tech. rept.
Chappell, David. 2010. Introducing Windows Communication Foundation. Tech. rept. January. Microsoft Corporation.
Chesbrough, Henry W. 2007. Why Companies Should Have Open
Business Models. MIT Sloan Management Review, 48(2), 22–28.
Christopher, Martin, Payne, Adrian, & Ballantyne, David. 1993. Relationship Marketing: Bringing quality, customer service and marketing
together. Butterworth-Heinemann.
Coase, Ronald H. 1937. The Nature of the Firm. Economica, 4(16),
386–405.

Bibliography

289

Crosby, Lawrence A., Evans, Kenneth R., & Cowles, Deborah. 1990.
Relationship Quality in Services Selling : An Interpersonal Influence Perspective. Journal of Marketing, 54(3), 68–81.
Cuganesan, Suresh. 2008. Calculating customer intimacy: accounting numbers in a sales and marketing department. Accounting,
Auditing & Accountability Journal, 21(1), 78–103.
Dalkir, Kimiz. 2011. Knowledge Management in Theory and Practice. 2
edn. The MIT Press.
Davenport, Thomas H. 2006. Competing on analytics. Harvard Business Review, 84(1), 98–107, 134.
Davenport, Thomas H., & Harris, Jeanne G. 2007. Competing on Analytics: The New Science of Winning. 1 edn. Harvard Business School
Press.
Day, George S. 2003. Creating a Superior Customer-Relating Capability. MIT Sloan Management Review.
Day, George S., & Montgomery, David B. 1999. Charting New Directions for Marketing. Journal of Marketing, 63(Special Issue 1999),
3–13.
De Choudhury, Munmun, Mason, Winter A., Hofman, Jake M., &
Watts, Duncan J. 2010. Inferring relevant social networks from
interpersonal communication. Proceedings of the 19th international
conference on World wide web - WWW ’10, 301.
De Jong, Ad, & De Ruyter, Ko. 2004. Adaptive versus Proactive
Behavior in Service Recovery: The Role of Self-Managing Teams.
Decision Sciences, 35(3), 457–491.
De Wulf, Kristof, Odekerken-Schröder, Gaby, & Iacobucci, Dawn.
2001. Investments in Consumer Relationships : A Cross-Country
and Cross-Industry Exploration. Journal of Marketing, 65(October),
33–50.

290

Bibliography

Den Butter, Frank A. G. 2010. Transaction Management : Value Creation
by Reducing Transaction Costs.
Deng, Shengliang, & Dart, Hack. 1994. Measuring Market Orientation: A Multi-factor, Multi-item Approach. Journal of Marketing
Management, 10(8), 725–742.
Dick, Alan S., & Basu, Kunal. 1994. Customer Loyalty: Toward an
Integrated Conceptual Framework. Journal of the Academy of Marketing Science, 22(2), 99–113.
Dickie, Jim. 2007. Fueling the crm engine. DestinationCRM.com.
Doherty, Neil F., & Lockett, Nigel J. 2004. Closing the Gap Between
the Expectations of Relationship Marketing and the Reality of ECRM. International Journal of E-Business Research, 3(2), I–VI.
Donaldson, Bill, & O’Toole, Tom. 2007. Strategic market relationships.
2nd edn. John Wiley & Sons.
Dunham, Margaret H. 2002. Data Mining: Introductory and Advanced
Topics. Prentice Hall.
Dyer, Jeffrey H. 1997. Effective interim collaboration: how firms minimize transaction costs and maximise transaction value. Strategic
Management Journal, 18(7), 535–556.
Dyer, Jeffrey H., & Chu, Wujin. 2003. The Role of Trustworthiness in
Reducing Transaction Costs and Improving Performance: Empirical Evidence from the United States, Japan, and Korea. Organization Science, 14(1), 57–68.
Elzinga, D.Jack, Horak, Tomas, Lee, Chung-Yee, & Bruner, Charles.
1995. Business process management: survey and methodology.
IEEE Transactions on Engineering Management, 42(2), 119–128.
Ernerst-Jones, Terry. 2005. Know how - Managing knowledge for competitive advantage. Tech. rept. The Economist.

Bibliography

291

Etzion, Opher, & Niblett, Peter. 2010. Event Processing in Action. Manning Publications.
Fayyad, Usama M., Piatetsky-Shapiro, Gregory, & Smyth, Padhraic.
1996a. From Data Mining to Knowledge Discovery: An Overview.
Chap. 1, page 611 of: Fayyad, Usama M., Piatetsky-Shapiro, Gregory, Smyth, Padhraic, & Uthurusamy, Ramasamy (eds), Advances
in Knowledge Discovery and Data Mining. AAAI Press / The MIT
Presse.
Fayyad, Usama M., Piatetsky-Shapiro, Gregory, & Smyth, Padhraic.
1996b. From Data Mining to Knowledge Discovery in Databases.
AI Magazine, 17(3), 37–54.
Fornell, Claes, Johnson, Michael D., Anderson, Eugene W., Cha, Jaesung, & Bryant, Barbara Everitt. 1996. The American Customer Satisfaction Index: Nature, Purpose, and Findings. Journal of Marketing, 60(4), 7.
Freeman, L. 1979. Centrality in social networks conceptual clarification. Social Networks, 1(3), 215–239.
Frese, Michael, Fay, Doris, Hilburger, Tanja, Leng, Karena, & Tag, Almut. 1997. The concept of personal initiative: Operationalization,
reliability and validity in two German samples. Journal of occupational and organizational psychology, 70(2), 139–161.
Gartner. 2010. Hype Cycle for Analytic Applications. Tech. rept. October.
Gebert, Henning, Geib, Malte, Kolbe, Lutz, & Brenner, Walter. 2003.
Knowledge-enabled customer relationship management: integrating customer relationship management and knowledge management concepts. Journal of Knowledge Management, 7(5), 107–123.
George, Michael L. 2003. Lean Six Sigma for Service : How to Use
Lean Speed and Six Sigma Quality to Improve Services and Transactions.
McGraw-Hill.

292

Bibliography

Gibbert, Michael, Leibold, Marius, & Probst, Gilbert. 2002. Five styles
of Customer Knowledge Management, and how smart companies
put them into action. European Management Journal, 20, 459–469.
Granovetter, Mark. 1985. Economic Action and Social Structure : The
Problem of Embeddedness. The American Journal of Sociology, 91(3),
481–510.
Grönroos, Christian. 1994. From Marketing Mix to Relationship
Marketing: Towards a Paradigm Shift in Marketing. Management
Decision, 32(2), 4–20.
Grönroos, Christian. 1997. Value-driven relational marketing: From
products to resources and competencies. Journal of Marketing Management, 13(5), 407–419.
Grönroos, Christian. 2000. Relationship Marketing - The Nordic
School Perspective. Chap. 4, pages 95–117 of: Sheth, Jagdish N.,
& Parvatiyar, Atul (eds), Handbook of Relationship Marketing, 1 edn.
Sage Publications, Inc.
Grönroos, Christian. 2007. Service Management and Marketing: Customer Management in Service Competition. 3 edn. Wiley.
Gruen, Thomas W. 1997. Relationship marketing: the route to marketing efficiency and effectiveness. Business Horizons, 40(6), 32–38.
Gummesson, Evert. 1995. Relationsmarknadsföring: Från 4P till 30R.
Gummesson, Evert. 1996. Relationship marketing and imaginary organizations: a synthesis. European Journal of Marketing, 30(2), 31–44.
Gummesson, Evert. 2008. Total Relationship Marketing. ButterworthHeinemann.
Gwinner, Kevin P., Bitner, Mary Jo, Brown, Stephen W., & Ajith, Kumar. 2005. Service Customization Through Employee Adaptiveness. Journal of Service Research, 8(2), 131–148.

Bibliography

293

Habryn, François, Blau, Benjamin, Satzger, Gerhard, & Kölmel, Bernhard. 2010. Towards a Model for Measuring Customer Intimacy in
B2B Services. Pages 1–14 of: Morin, Jean-Henry, Ralyté, Jolita, &
Snene, Mehdi (eds), Exploring Services Science. Springer Berlin Heidelberg.
Hair, Joseph F., Black, William C., Babin, Barry J., & Anderson,
Rolph E. 2010. Multivariate Data Analysis (7th Edition).
Hakansson, Hakan, & Snehota, Ivan. 2000. The IMP Perspective: Assets and Liabilities of Business Relationships. Chap. 3, pages 69–93
of: Sheth, Jagdish N., & Parvatiyar, Atul (eds), Handbook of Relationship Marketing. Sage Publications, Inc.
Hakansson, Hakan, Ford, David I., Gadde, Lars-Erik, Snehota, Ivan,
& Waluszewski, Alexandra. 2009. Business in Networks. Wiley.
Hambrick, Donald C. 1980.
Operationalizing the Concept of
Business-Level Strategy in Research. Academy of Management Journal, 5(4), 567–575.
Harding, Ford. 2004. Cross Selling or Cross Purposes ?
Business Review Case Study.

Harvard

Hasan, Matt. 2003. Ensure success of CRM with a change in mindset.
Marketing News, 37(8), 16.
Hennig-Thurau, Thorsten, Gwinner, Kevin P., & Gremler, Dwayne D.
2002. Understanding Relationship Marketing Outcomes: An Integration of Relational Benefits and Relationship Quality. Journal of
Service Research, 4(3), 230–247.
Heskett, James L, Jones, Thomas O, Loveman, Gary W, Sasser,
W Earl, & Schlesinger, Leonard A. 1994. Putting the Service-Profit
Chain to Work. Harvard Business Review, 72(2), 164–170.
Hodnick, Mike. 2009. Connectable Lines and Shapes in Silverlight.

294

Bibliography

Holmlund, Maria. 2004. Analyzing business relationships and distinguishing different interaction levels. Industrial Marketing Management, 33(4), 279–287.
Holt, Sue, & McDonald, Malcolm. 2000. Managing Global Networks
: The Role of the Global Account Manager. Pages 1–16 of: 16th IMP
Conference.
Homburg, Christian, & Jensen, Ove. 2004. Kundenbindung im Industriegütergeschäft. Universität Mannheim - Institut für Marktorientierte Unternehmensführung; Auflage: 1., Aufl.
Husted, Bryan W. 1994. Transaction Costs, Norms, and Social Networks: A Preliminary Study of Cooperation in Industrial BuyerSeller Relationships in the United States and Mexico. Business &
Society, 33(1), 30–57.
Hutt, Michael D., & Walker, Beth A. 2006. A network perspective
of account manager performance. Journal of Business & Industrial
Marketing, 21(7), 466–473.
IBM Institute for Business Value. 2010. Capitalizing on Complexity Insight from the Global CEO Study. Tech. rept.
IBM Institute for Business Value. 2011. Insights from the Global Chief
Marketing Officer Survey - From Stretched to Strenghened. Tech. rept.
Industry Directions, & IBM. 2006. Increasing Customer Intimacy, ValueAdd and Profits Through Expanded Services. Tech. rept.
Jamieson, Susan. 2004. Likert scales: how to (ab)use them. Medical
education, 38(12), 1217–8.
Jaworski, Bernard J., & Kohli, Ajay K. 1993. Market Orientation:
Antecedents and Consequences. The Journal of Marketing, 57(3),
53–70.
Jayachandran, Satish, Hewett, Kelly, & Kaufman, Peter. 2004. Customer Response Capability in a Sense-and-Respond Era: The Role

Bibliography

295

of Customer Knowledge Process. Journal of the Academy of Marketing Science, 32(3), 219–233.
John, George H. 1995. Robust Decision Trees : Removing Outliers
from Databases. Pages 174–179 of: Fayyad, Usama M., & Uthurusamy, R. (eds), Proceedings of the First International Conference on
Knowledge Discovery and Data Mining. Menlo Park: AAAI Press.
Joshi, Ashwin W., & Sharma, Sanjay. 2004. Customer Knowledge
Development : Antecedents and Impact on New Product Performance. Journal of Marketing, 68(October), 47–59.
Kale, Sudhir H. 2004. CRM Failure and the Seven Deadly Sins. Marketing Management, 13(5), 42–46.
Kalousis, Alexandros, Gama, João, & Hilario, Melanie. 2004. On Data
and Algorithms: Understanding Inductive Performance. Machine
Learning, 54(3), 275–312.
Kamakura, Wagner A., Ramaswamy, Sridhar N., & Srivastava, Rajendra K. 1991. Applying latent trait analysis in the evaluation of
prospects for cross-selling of financial services. International Journal
of Research in Marketing, 8(4), 329–349.
Kamakura, Wagner A., Wedel, Michel, De Rosa, Fernando, & Mazzon, Jose Afonso. 2003. Cross-selling through database marketing:
a mixed data factor analyzer for data augmentation and prediction.
International Journal of Research in Marketing, 20(1), 45–65.
Kaplan, Robert S. 2005. A Balanced Scorecard Approach To Measure
Customer Profitability.
Kaplan, Robert S., & Anderson, Steven R. 2007. Time-Driven ActivityBased Costing: A Simpler and More Powerful Path to Higher Profits.
Harvard Business School Press.
Kirkman, Bradley L., Rosen, Benson, Tesluk, Paul E., & Gibson,
Cristina B. 2004. The Impact of Team Empowerment on Virtual
Team Performance: The Moderating Role of Face-to-Face Interaction. The Academy of Management Journal, 47(2), 175–192.

296

Bibliography

Kiron, David, Shockley, Rebecca, Kruschwitz, Nina, Finch, Glenn, &
Haydock, Michael. 2011. Analytics : The widening divide. Tech. rept.
IBM Institute for Business Value.
Kiss, Christine. 2007. Leveraging social networks for analytical customer
relationship management. Ph.D. thesis.
Knoke, David, & Yang, Sang. 2008. Social Network Analysis. 2 edn.
Sage Publications, Inc.
Kohavi, Ron. 1995. A Study of Cross-Validation and Bootstrap for
Accuracy Estimation and Model Selection. Pages 1137–1143 of: Proceedings of the 14th International Joint Conference on Artificial Intelligence.
Kohli, Ajay K., & Jaworski, Bernard J. 1990. Market Orientation : The
Construct, Research Propositions Managerial Implications. Journal
of Marketing, 54(2), 1–18.
Kong, Raymond, & Mayo, Michael C. 1993. Measuring Service Quality in the Business-to-business Context. Journal of Business and Industrial Marketing, 8(2), 5–15.
Koschützki, Dirk, Lehmann, Katharina Anna, Peeters, Leon, Richter,
Stefan, Tenfelde-Podehl, Dagmar, & Zlotowski, Oliver. 2005. Centrality Indices. Chap. 3, pages 16–61 of: Brandes, Ulrik, & Erlebach, Thomas (eds), Network Analysis: Methodological Foundations.
Springer.
Kumar, Ashok. 2005. Mass Customization: Metrics and Modularity.
International Journal of Flexible Manufacturing Systems, 16(4), 287–
311.
Lages, Carmen, Lages, Cristiana Raque, & Lages, Luis Filipe. 2005.
The RELQUAL scale: a measure of relationship quality in export
market ventures. Journal of Business Research, 58(8), 1040–1048.
Lam, Shun Y., Shankar, Venkatesh, Erramilli, M. Krishna, & Murthy,
Bvsan. 2004. Customer Value, Satisfaction, Loyalty, and Switching

Bibliography

297

Costs: An Illustration From a Business-to-Business Service Context. Journal of the Academy of Marketing Science, 32(3), 293–311.
Lam, Wai, Keung, Chi-kin, & Liu, Danyu. 2002. Discovering useful
concept prototypes for classification based on filtering and abstraction. IEEE Transactions on Pattern Analysis and Machine Intelligence,
24(8), 1075–1090.
Leek, Sheena, Turnbull, Peter W., & Naudé, Pete. 2001. Interactions
, Relationships and Networks - Past , Present and Future. Pages
1–13 of: 17th IMP-conference, vol. 1.
Liang, Ting-peng. 2009. Information Technology for Customer Intimacy : A Niche for Research in the Internet Age. Pacific Asia
Journal of the Association for Information Systems, 1(3), 1–4.
Liljander, Veronika, & Strandvik, Tore. 1995. The Nature of Relationships in Services. Advances in Service Marketing and Management, 4,
143.
Lovelock, Christopher H., & Wirtz, Jochen. 2007. Services Marketing.
6., inter. edn. Pearson Prentice Hall.
Magnusson, Peter R., Matthing, Jonas, & Kristensson, Per. 2003. Managing User Involvement in Service Innovation: Experiments with
Innovating End Users. Journal of Service Research, 6(2), 111–124.
Malms, Oliver, & Schmitz, Christian. 2011. Cross-Divisional Orientation: Antecedents and Effects on Cross-Selling Success. Journal of
Business-to-Business Marketing, 18(3), 253–275.
Manasco, Britton. 2000.
APQC, 66–70.

Capitalizing on Customer Knowledge.

Maoz, Michael, Davies, Jim, Thompson, Ed, Dunne, Michael, &
Sarner, Adam. 2010. The Concise Social CRM Vendor Guide , 2011.
Tech. rept. October 2010. Gartner, Inc.

298

Bibliography

McDonald, Malcolm, Millman, Tony, & Rogers, Beth. 1997. Key account management: Theory, practice and challenges. Journal of
Marketing Management, 13(8), 737–757.
Medlin, Christopher John, & Törnroos, Jan-Ake. 2006. Inter-firm Interaction from a Human Perspective. In: 22nd IMP-conference.
Michie, Donald, Spiegelhalter, David J., & Taylor, Charles C. 1994.
Machine Learning, Neural, and Statistical Classification. Journal of
the American Statistical Association, 91(433), 436– 438.
Miller, Delbert C., & Salkind, Neil J. 2002. Handbook of Research Design
and Social Measurement. 6th edn. Sage Publications, Inc.
Mitchell, James C. 1969. Social Network in Urban Situations: analyses of
personal relationships in Central African towns.
Mitchell, Tom M. 1999. Machine Learning and Data Mining Over the
past. Communications of the ACM, 42(11).
Mohr, Jakki, & Nevin, John R. 1990. Communication Strategies in
Marketing Channels : A Theoretical Perspective. The Journal of
Marketing, 54(4), 36–51.
Morgan, Robert M., & Hunt, Shelby D. 1994. The Commitment-Trust
Theory of Relationship Marketing. Journal of Marketing, 58(3), 20–
38.
Narver, John C., & Slater, Stanley F. 1990. The Effect of a Market Orientation on a Market Business Profitability. The Journal of Marketing,
54(4), 20–35.
Nash, Debbie, & Sterna-Karwat, Alicia. 1996. An application of DEA
to measure branch cross selling efficiency. Computers & Operations
Research, 23(4), 385–392.
Ndubisi, Nelson Oly. 2007. Relationship marketing and customer
loyalty. Marketing Intelligence & Planning, 25(1), 98–106.

Bibliography

299

Neely, Andy. 2009. Exploring the financial consequences of the servitization of manufacturing. Operations Management Research, 1(2),
103–118.
Newman, Mark E.J. 2004. Analysis of weighted networks. Physical
Review E, 70(5), 1–9.
Nezlek, John B. 2003. Using Multilevel Random Coefficient Modeling to Analyze Social Interaction Diary Data. Journal of Social and
Personal Relationships, 20(4), 437–469.
Nisbet, Robert, Elder, John, & Miner, Gary. 2009. Handbook of Statistical Analysis and Data Mining Applications. Academic Press.
Niven, Paul R. 2002. Balanced Scorecard Step-by-Step: Maximizing Performance and Maintaining Results. Wiley.
Noorderhaven, Niels, & Harzing, Anne-Wil. 2009. Knowledgesharing and social interaction within MNEs. Journal of International
Business Studies, 40(5), 719–741.
Odekerken-Schröder, Gaby, De Wulf, Kristof, & Schumacher, Patrick.
2003. Strengthening outcomes of retailer-consumer relationships.
The dual impact of relationship marketing tactics and consumer
personality. Journal of Business Research, 56(3), 177–190.
Oliva, Rogelio, & Kallenberg, Robert. 2003. Managing the transition
from products to services. International Journal of Service Industry
Management, 14(2), 160–172.
Oliver, Richard L. 1999. Whence Consumer Loyalty? Journal of Marketing, 63(1999), 33.
Ostaszewski, Adam. 1990. Advanced mathematical methods. Cambridge
University Press.
Palmatier, Robert W., Dant, Rajiv P., Grewal, Dhruv, & Evans, Kenneth R. 2006. Factors Influencing the Effectiveness of Relationship
Marketing: A Meta-Analysis. Journal of Marketing, 70(October),
136–153.

300

Bibliography

Palmatier, Robert W., Scheer, Lisa K., & Steenkamp, JanBenedict E.M. 2007. Customer Loyalty to Whom? Managing the
Benefits and Risks of Salesperson-Owned Loyalty. Journal of Marketing Research, 44(2), 185–199.
Parvatiyar, Atul, & Sheth, Jagdish N. 2000. The Domain and Conceptual Foundations of Relationship Marketing. Chap. 1, pages 3–38 of:
Sheth, Jagdish N., & Parvatiyar, Atul (eds), Handbook of Relationship
Marketing, 1 edn. Sage Publications, Inc.
Parvatiyar, Atul, & Sheth, Jagdish N. 2001. Conceptual Framework of
Customer Relationship Management. Chap. 1, pages 3–25 of: Sheth,
Jagdish N., Parvatiyar, Atul, & Shainesh, G (eds), Customer relationship management: emerging concepts, tools, and applications, 1 edn.
Tata McGraw-Hill.
Patton, Michael Q. 2002. Qualitative research and evaluation methods. 3
edn. Sage Publications, Inc.
Payne, Adrian. 2005. Handbook of CRM: Achieving Excellence through
Customer Management. Butterworth-Heinemann.
Payne, Adrian, & Frow, Pennie. 2005. A Strategic Framework for
Customer Relationship Management. Journal of Marketing, 69(4),
167–176.
Plinke, Wulff. 1997. Grundlagen des Geschäftsbeziehungsmanagement. Chap. 1, pages 1–62 of: Kleinaltenkamp, Michael, & Plinke,
Wulff (eds), Geschäftsbeziehungsmanagement. Springer.
Porter, Michael E. 2004. Competitive Strategy. Free Press.
Potgieter, A, & Roodt, G. 2004. Measuring a customer intimacy culture in a value discipline context. South African Journal of Human
Resource Management, 2(3), 25–31.
Press, S. James. 2003. The Role of Bayesian and Frequentist Multivariate Modeling in Statistical Data Mining. Chap. 1, page 624 of:
Bozdogan, Hamparsum (ed), Statistical Data Mining and Knowledge
Discovery. Chapman and Hall/CRC.

Bibliography

301

Quinlan, John R. 1986. Induction of decision trees. Machine Learning,
1(1), 81–106.
Reichheld, Frederick F. 2003. The one number you need to grow.
Harvard Business Review, 81(12), 46–54, 124.
Reichheld, Frederick F., & Sasser, W. Earl. 1990. Zero defections:
quality comes to services. Harvard Business Review, 68(5), 105–111.
Reichheld, Frederick F., & Teal, Thomas. 2001. The loyalty effect: the
hidden force behind growth, profits, and lasting value. Harvard Business Press.
Reinartz, Werner J., & Kumar, V. 2000. On the Profitability of LongLife Customers in a Noncontractual Setting: An Empirical Investigation and Implications for Marketing. Journal of Marketing, 64(4),
17–35.
Reinartz, Werner J., Krafft, Manfred, & Hoyer, Wayne D. 2004. The
Customer Relationship Management Process: Its Measurement
and Impact on Performance. Journal of Marketing Research, 41(3),
293–305.
Richard, James E. 2008. The Impact of Customer Relationship Management (CRM) Technology on Business-to-Business Customer Relationships. Ph.D. thesis, Victoria University of Wellington.
Richards, Keith A., & Jones, Eli. 2008. Customer relationship management: Finding value drivers. Industrial Marketing Management,
37(2), 120–130.
Richardson, Leonard, & Ruby, Sam. 2007. Restful Web Services. 1st
edn. O’Reilly Media, Inc, USA.
Roberts, Keith, Varki, Sajeev, & Brodie, Rod. 2003. Measuring the
quality of relationships in consumer services: an empirical study.
European Journal of Marketing, 37(1/2), 169–196.
Robinson, John P., Shaver, Phillip R., & Wrightsman, Lawrence S.
1991. Criteria for Scale Selection and Evaluation. Chap. 1, pages

302

Bibliography

1–16 of: Robinson, John P, Shaver, Phillip R, & Wrightsman,
Lawrence S (eds), Measures of personality and social psychological attitudes. Elsevier.
Rodrigez Cano, Cynthia, Carillat, Francois A., & Jaramillo, Fernando.
2004. A meta-analysis of the relationship between market orientation and business performance: evidence from five continents.
International Journal of Research in Marketing, 21(2), 179–200.
Rust, Roland T., Lemon, Katherine N., & Zeithaml, Valarie A. 2004.
Return on Marketing: Using Customer Equity to Focus Marketing
Strategy. Journal of Marketing, 68(1), 109–127.
Ryals, Lynette, & Knox, Simon. 2001. Cross-functional issues in the
implementation of relationship marketing through customer relationship management. European Management Journal, 19(5), 534–
542.
Sabidussi, Gert. 1966. The centrality index of a graph. Psychometrika,
31(4), 581–603.
Sackmann, Stefan, Kundisch, Dennis, & Ruch, Markus. 2008.
CRM, Kundenbewertung und integrierte Risk-Return-Steuerung
- Zum Status quo des betrieblichen Einsatzes. HMD Praxis der
Wirtschaftsinformatik, 44(259), 21–32.
Sako, Mari. 1992. Prices, quality, and trust: inter-firm relations in Britain
and Japan.
Sandberg, Birgitta. 2007. Customer-related proactiveness in the radical innovation development process. European Journal of Innovation
Management, 10(2), 252–267.
Satzger, Gerhard, & Neus, Andreas. 2010. Principles of Collaborative
Innovation: Implementation of Open and Peer-to-Peer Innovation
Approaches. Pages 219–238 of: Gerybadze, Alexander, Hommel,
Ulrich, Reiners, Hans W., & Thomaschewski, Dieter (eds), Innovation and International Corporate Growth. Berlin, Heidelberg: Springer
Berlin Heidelberg.

Bibliography

303

Shapiro, Benson P., & Wyman, John. 1981. New Ways to Reach Your
Customers. Harvard Business Review, 59(6), 103–110.
Sharma, Neeru, & Patterson, Paul G. 1999. The impact of communication effectiveness and service quality on relationship commitment in consumer, professional services. Journal of Services Marketing, 13(2), 151–170.
Sheth, Jagdish N., & Parvatiyar, Atul. 2000. The evolution of relationship marketing. Chap. 5, pages 119–145 of: Sheth, Jagdish N.,
& Parvatiyar, Atul (eds), Handbook of Relationship Marketing, vol. 4.
Sage Publications, Inc.
Sheth, Jagdish N., & Sharma, Arun. 2008. The impact of the product
to service shift in industrial markets and the evolution of the sales
organization. Industrial Marketing Management, 37(3), 260–269.
Silveira, Da Giovanni, Borenstein, Denis, & Fogliatto, Flávio S. 2001.
Mass customization: Literature review and research directions. International Journal of Production Economics, 72(1), 1–13.
Sommerville, Ian. 2007. Software Engineering. 8th edn. Pearson Education.
Stern, Barbara B. 1997. Advertising Intimacy : Relationship Marketing and the Services Consumer. Journal of Advertizing, 26(4), 7–19.
Storbacka, Kaj, Strandvik, Tore, & Grönroos, Christian. 1994. Managing Customer Relationships for Profit: The Dynamics of Relationship Quality. International Journal of Service Industry Management,
5(5), 21–38.
Tan, Pang-Ning, Steinbach, Michael, & Kumar, Vipin. 2006. Introduction to Data Mining. Addison Wesley.
Treacy, Michael, & Wiersema, Frederik D. 1993. Customer intimacy
and other value disciplines. Harvard Business Review.
Treacy, Michael, & Wiersema, Frederik D. 1997. The discipline of market
leaders. Basic Books.

304

Bibliography

Tsvetovat, Maksim, Reminga, Jeff, & Carley, Kathleen M. 2004.
DyNetML : Interchange Format for Rich Social Network Data. Tech.
rept. February. Carnegie Mellon University - School of Computer
Science, Pittsburg.
Tu, Y. L., Xie, S. Q., & Fung, Richard Y. K. 2007. Product Development Cost Estimation in Mass Customization. IEEE Transactions on
Engineering Management, 54(1), 29–40.
Tuominen, Matti, Rajala, Arto, & Mo, Kristian. 2004. Market-driving
versus market-driven: Divergent roles of market orientation in
business relationships. Industrial Marketing Management, 33(3), 207–
217.
Turban, Efraim, Sharda, Ramesh, & Delen, Dursun. 2011. Decision
Support and Business Intelligence Systems (International Edition). 9th
edn. Pearson Education.
Vandermerwe, Sandra. 1988. Servitization of business: Adding value
by adding services. European Management Journal, 6(4), 314–324.
Varadarajan, P. Rajan, & Cunningham, Margaret H. 1995. Strategic
Alliances: A Synthesis of Conceptual Foundations. Journal of the
Academy of Marketing Science, 23(4), 282–296.
Varadarajan, P. Rajan, & Rajaratnam, Daniel. 1986. Symbiotic Marketing Revisited. Journal of Marketing, 50(1), 7.
Vargo, Stephen L., & Lusch, Robert F. 2004a. Evolving to a New
Dominant Logic for Marketing. Journal of Marketing, 68(1), 1–17.
Vargo, Stephen L., & Lusch, Robert F. 2004b. The Four Service Marketing Myths: Remnants of a Goods-Based, Manufacturing Model.
Journal of Service Research, 6(4), 324–335.
Vargo, Stephen L., & Lusch, Robert F. 2006. Service-Dominant Logic What It Is, What It Is Not, What It Might Be. Chap. 3, pages 43–56 of:
Vargo, Stephen L., & Lusch, Robert F. (eds), The Service-Dominant
Logic of Marketing, 1 edn. M.E.Sharpe.

Bibliography

305

Vargo, Stephen L., & Lusch, Robert F. 2008a. Service-dominant logic:
continuing the evolution. Journal of the Academy of Marketing Science, 36(1), 1–10.
Vargo, Stephen L., & Lusch, Robert F. 2008b. Why "service"? Journal
of the Academy of Marketing Science, 36(1), 25–38.
Venkatesan, Rajkumar, & Kumar, V. 2004. A Customer Lifetime Value
Framework for Customer Selection and Resource Allocation Strategy. Journal of Marketing, 68(4), 106–125.
Wallenburg, Carl Marcus. 2009. Innovation in Logistics Outsourcing Relationships: Proactive Improvement by Logistics Service
Providers as a Driver of Customer Loyalty. Journal of Supply Chain
Management, 45(2), 75–93.
Wallenburg, Carl Marcus, Knemeyer, A. Michael, Goldsby, Thomas J.,
& Cahill, David L. 2010. Developing a scale for proactive improvement within logistics outsourcing relationships. The International
Journal of Logistics Management, 21(1), 5–21.
Wasserman, Stanley, & Faust, Katherine. 1994. Social Network Analysis: Methods and Applications. 1 edn. Cambridge University Press.
Weill, Peter, & Ross, Jeanne W. 2004. IT governance: how top performers manage IT decision rights for superior results. Harvard Business
School Press.
Wengler, Stefan. 2006. Key account management in business-to-business
markets. Ph.D. thesis.
Wengler, Stefan, Ehret, Michael, & Saab, Samy. 2006. Implementation of Key Account Management: Who, why, and how?: An exploratory study on the current implementation of Key Account
Management programs. Industrial Marketing Management, 35(1),
112–103.
Williamson, Oliver E. 1979. Transaction-Cost Economics: The Governance of Contractual Relations. Journal of Law and Economics, 22(2),
233–261.

306

Bibliography

Wind, Jerry, & Rangaswamy, Arvind. 2001. Customerization: The
next revolution in mass customization. Journal of Interactive Marketing, 15(1), 13–32.
Witten, Ian H., Eibe, Frank, & Hall, Mark A. 2011. Data Mining: Practical Machine Learning Tools and Techniques. 3 edn. Morgan Kaufmann.
Wu, Lynn, Lin, Ching-yung, Stern, N Y U, & Brynjolfsson, Erik. 2009.
Value of Social Network – A Large-Scale Analysis on Network
Structure Impact to Financial Revenue of Information Technology
Consultants. In: Winter Information Systems Conference.
Wuyts, Stefan, & Geyskens, Inge. 2005. The Formation of BuyerSupplier Relationships: Detailed Contract Drafting and Close Partner Selection. Journal of Marketing, 69(4), 103–117.
Yim, Chi Kin, Tse, David K, & Chan, Kimmy Wa. 2008. Strengthening Customer Loyalty Through Intimacy and Passion: Roles of
Customer-Firm Affection and Customer-Staff Relationships in Services. Journal of Marketing Research, 45(6), 741–756.
Zablah, Alex, Bellenger, Danny N., & Johnston, Wesley J. 2004.
An evaluation of divergent perspectives on customer relationship
management: Towards a common understanding of an emerging
phenomenon. Industrial Marketing Management, 33(6), 475–489.
Zack, Michael, McKeen, James, & Singh, Satyendra. 2009. Knowledge
management and organizational performance: an exploratory analysis. Journal of Knowledge Management, 13(6), 392–409.
Zajac, Edward J., & Olsen, Cyrus P. 1993. From Transaction Cost
to Transactional Value Analysis: Implications for the Study of Interorganizational Strategies. Journal of Management Studies, 30(1),
131–145.

Appendix
A. Questionnaire Customer Intimacy
This appendix presents the questionnaire conceived in the scope
of this thesis in order to perform the empirical assessment of the
customer intimacy components. This questionnaire consists of four
different parts:
1. Introduction: This section introduces the scope of the survey
2. Acquired Customer Intimacy – Organization Level: In this
part of the questionnaire, the acquired customer intimacy components at the organizational level are empirically assessed on
Likert-type scales with a set of four items.
3. Acquired Customer Intimacy – Individual Level: In this part
of the questionnaire, the acquired customer intimacy components at the individual level are empirically assessed with a set
of four items.
4. Work Environment: Finally, in this part of the questionnaire,
the respondents are asked to provide further information on
their work environment. This part consists of 11 items.

A.1. English Version
In this section, the English version of the questionnaire is presented

Appendix

in partnership with

308

Figure A.1.: Customer Intimacy Questionnaire: Introduction

309

1

2

3

4

5

6

7

A. Questionnaire Customer Intimacy

Figure A.2.: Customer Intimacy Questionnaire: Acquired Customer
Intimacy at the Organizational Level

Appendix

1

2

3

4

5

6

7

310

Figure A.3.: Customer Intimacy Questionnaire: Acquired Customer
Intimacy at the Individual Level

311

1

2

3

4

5

6

7

A. Questionnaire Customer Intimacy

Figure A.4.: Customer Intimacy Questionnaire: Work Environment

312

Appendix

A.2. German Version
Since the respondents of the survey are from Germany, the customer
intimacy questionnaire has been translated in the German language.
This section presents this translated questionnaire.

313

in Kooperation mit

A. Questionnaire Customer Intimacy

Figure A.5.: Customer
(German)

Intimacy

Questionnaire:

Introduction

Appendix

1

2

3

4

5

6

7

314

Figure A.6.: Customer Intimacy Questionnaire: Acquired Customer
Intimacy at the Organizational Level (German)

315

1

2

3

4

5

6

7

A. Questionnaire Customer Intimacy

Figure A.7.: Customer Intimacy Questionnaire: Acquired Customer
Intimacy at the Individual Level (German)

Appendix

1

2

3

4

5

6

7

316

Figure A.8.: Customer Intimacy Questionnaire: Work Environment
(German)

B. Machine Learning Algorithms Settings

317

B. Machine Learning Algorithms Settings
This appendix consists of five tables. The first four tables describe
the considered parameters for configuring the machine learning algorithms used in this thesis and elaborated in chapter 7: the decision tree C4.5, the multilayer perceptron with backpropagation neural network, k-nearest neighbour, and the support vector machine.
The description of the individual options is derived from the Weka
documentation.1 The exact list of parameter combinations tested in
this project is available upon request from the author. Finally, table B.5 details the number of configurations tested for each machine
learning algorithm and for each predicted variable.

1

Further details are available at http://www.cs.waikato.ac.nz/ml/weka/ (accessed
on 1.12.2011).

318

Appendix

Table B.1.: Configuration Settings of the Decision Tree C4.5
Option
binarySplit

Considered Values

Description

True / False

Whether data splits on nominal attributes are binary or
not

From 0.1 to 0.8 with inDegree of pruning of the tree
crements of 0.1
From 2 to 10 with incre- Minimum number of objects
minNumObj
ments of 1
per terminal leaf
Whether to use reduced error
reducedErrorPruningTrue / False
pruning instead of C4.5 error
pruning or not
Amount of data used for
From 2 to 5 with incre- reduced-error raising prunnumFolds
ments of 1
ing (if reducedErrorPruning
is set to true)
Whether to use the subtree
subTreeRaising
True / False
raising operation or not during the pruning task
Whether to use the Laplace
function when counting the
useLaplace
True / False
the number of instances at a
node
Whether to perform the
unpruned
True / False
pruning task or not
confidenceFactor

B. Machine Learning Algorithms Settings

Table B.2.: Configuration
Algorithm
Option
KNN
crossValidate

distanceWeighting

meanSquared

NearestNeighbor
SearchAlgorithm

Settings

Considered Values

of

319

the

k-nearest

Neighbor

Description

From 1 to 10 with increNumber of neighbors to use
ments of 1
Use
hold-one-out
crossTrue / False
validation to select the best k
value
No Distance Weighting
Determines the distance
Weight by 1/distance
weighting method
Weight by 1 - distance
Determines whether to use
True / False
the mean squared error or the
mean absolute error
LinearNNSearch
The nearest neighbour
BallTree
search algorithm to use
CoverTree
KDTree

Table B.3.: Configuration Settings of the Support Vector Machine
Algorithm
Option

Considered Values

buildLogisticModels False

Description
Whether to fit logistic models
to the outputs

0.5 to 2.5 with increComplexity Parameter C
ments of 0.1
The epsilon for round-off erEpsilon
1.0.E−12
ror
Polykernel
Puk
Kernel
The Kernel to use
RBFKernel
NormalizedPolyKernel
toleranceParameter 0.0010
The tolerance parameter
c

320

Appendix

Table B.4.: Configuration Settings of the Multilayer Perceptron with
Backpropagration
Option

Considered Values

decay

True / False

learningRate

from 0.0 to 1.0

Momentum

from 0.0 to 1.0

NominalToBinary

True / False

Reset

True / False

hiddenLayer

a

autobuild

True

trainingTime

1/500

validationSetTime

0

Description
Decreases the learning rate if
set to true
The amount the weights are
updated
Momentum applied to the
weights during updating
Can improve performance if
the data set contains binary
attributes
Determines the number of
hidden layers automatically
Adds and connects up hidden network automatically
The number of epochs to
train through.
No validation set will be used
and instead the network will
train for the specified number of epochs.

B. Machine Learning Algorithms Settings

321

Table B.5.: Number of Tested Configurations of the Machine Learning Algorithms to Predict the Customer Intimacy Values
Amount of Tested Configurations
C4.5

k-NN

SVM

NNBP

Total

Knowledge High

96

36

41

71

244

Knowledge Very High

74

29

41

86

230

Relationship High

52

29

46

67

194

Relationship Very
High

54

36

46

56

192

Knowledge High

27

43

46

57

173

Knowledge Very High

27

31

46

62

166

Relationship High

35

31

46

68

180

Relationship Very
High

27

31

46

62

166

392

266

358

529

1545

Individual Level

Organizational Level

Total

322

Appendix
Notes
Output Created

25-Mar-2011 09:41:21

C. Acquired Customer Intimacy at the
Individual Level
Comments
Input

Data

C:\Dokumente und
Einstellungen\Administrator\Desktop\
CIG_model_
calibration\Preprocessings\Further_
Preprocessings\Further_
preprocessings.sav

This appendix provides further details on the metrics calibration
Dataset
DataSet1
which is developed inActive
chapter
7
to
assess
the acquired customer inFilter
<none>
timacy components at
the individual<none>
level. Figure C.1 shows the
Weight
Split File of the summated
<none>
Crombach’s Alpha values
scales Knowledge and ReN of Rows in Working
117
lationship at the individual
Data File level. Table C.2 details the achieved caliMatrix Input
bration results to predict
the variable Knowledge High with the deciMissing Value Handling
Definition of Missing
User-defined missing values are
sion tree C4.5 algorithm. It can be observed
that 52 models have been
treated as missing.
Cases
Used
Statistics
based on all cases
with best results
created and tested, the
model
number
40 are
obtaining
the
valid data for all variables in the
procedure.
and being therefore chosen. Tables C.2,
C.3, C.4 and C.5 detail the
Syntax
RELIABILITY
/VARIABLES=Question21
best calibration settings of the four considered
machine learning alQuestion22
/SCALE('ALL
VARIABLES')
ALL
gorithms to assess the variables Knowledge
High, Knowledge
Very High,
/MODEL=ALPHA.
Resources
Time
00 00:00:00.016
Relationship
High, andProcessor
Relationship
Very High at the
individual level.
00 00:00:00.016
Scale: Relationship

Elapsed Time

Case Processing Summary

Case Processing Summary
N
Cases

Valid
Excluded

a

Total

N

%

117

100.0

0

.0

117

100.0

Cases

Valid

%

104
a

Excluded
Total

.0

104

100.0

a. Listwise deletion based on all
variables in the procedure.

a. Listwise deletion based on all
variables in the procedure.

Reliability Statistics

Reliability Statistics

Cronbach's
Alpha
.912

N of Items
2

(a) Scale Knowledge

Cronbach's
Alpha
.940

100.0

0

N of Items
2

(b) Scale Relationship

Figure C.1.: Crombach’s Alpha of the Scales Knowledge and Relationship at the Individual Level

Page 1

C. Acquired Customer Intimacy at the Individual Level

323

Table C.1.: Prediction of the Variable Knowledge High: Detailed Perfor-

Model Number

Binary Split

Confidence Factor

MinNumObj

NumFolds

Reduced-Error-Pruning

SubTree-Raising

Unpruned

Use-Laplace

Success Rate (%)

Precision (%)

Recall (%)

F-Measure (%)

Kappa Statistic (%)

mance Results of the Decision Tree C4.5

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0

0.25
0.25
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25

2
2
2
2
2
2
2
2
2
2
1
3
4
5
6
7
8
9
10
20
30
40
50
2
2
2
2
2
2
2

3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
2
3
4
5
3
3
3

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
0
0
0

1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0
1
1

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1

77.1
77.1
77.45
77.62
77.36
77.11
76.01
75.25
75.25
75.25
77.36
78.06
77.72
77.87
77.2
75.33
74.82
74.8
75.08
76.72
76.72
78.41
72.73
76.64
77.58
76.24
76.7
77.45
75.67
77.1

0.83
0.83
0.85
0.84
0.83
0.83
0.82
0.83
0.83
0.83
0.84
0.84
0.84
0.84
0.82
0.8
0.79
0.8
0.82
0.87
0.88
0.85
0.72
0.85
0.86
0.85
0.85
0.84
0.84
0.83

0.66
0.66
0.67
0.67
0.67
0.67
0.65
0.61
0.61
0.61
0.67
0.69
0.68
0.69
0.69
0.67
0.67
0.65
0.64
0.62
0.61
0.68
0.73
0.64
0.65
0.63
0.63
0.67
0.62
0.66

0.72
0.72
0.72
0.73
0.72
0.72
0.7
0.69
0.69
0.69
0.72
0.74
0.74
0.74
0.73
0.71
0.7
0.7
0.69
0.7
0.7
0.74
0.72
0.71
0.72
0.7
0.7
0.73
0.7
0.72

0.54
0.54
0.54
0.55
0.54
0.54
0.51
0.5
0.5
0.5
0.54
0.56
0.55
0.55
0.54
0.5
0.49
0.49
0.5
0.53
0.53
0.56
0.45
0.53
0.55
0.52
0.53
0.55
0.51
0.54

324

Appendix

Model Number

Binary Split

Confidence Factor

MinNumObj

NumFolds

Reduced-Error-Pruning

SubTree-Raising

Unpruned

Use-Laplace

Success Rate (%)

Precision (%)

Recall (%)

F-Measure (%)

Kappa Statistic (%)

Continued

31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0

0.2
0.2
0.2
0.2
0.2
0.2
0.2
0.1
0.3
0.4
0.5
0.6
0.7
0.8
0.2
0.4
0.4
0.4
0.4
0.4
0.4
0.4

1
2
4
5
6
7
8
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3

3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
2
3
4
5
3
3
3

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
0
0
0

1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0
1
1

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1

77.36
77.62
77.64
77.7
77.2
75.06
74.29
77.47
78.06
78.39
77.8
74.27
74.27
74.27
78.06
76.2
77.41
77.18
76.62
78.23
74.6
78.39

0.84
0.84
0.84
0.84
0.82
0.8
0.79
0.84
0.84
0.84
0.83
0.8
0.8
0.8
0.84
0.84
0.86
0.86
0.85
0.84
0.81
0.84

0.67
0.67
0.68
0.69
0.69
0.66
0.65
0.68
0.69
0.7
0.68
0.62
0.62
0.62
0.69
0.64
0.65
0.64
0.63
0.69
0.63
0.7

0.72
0.73
0.73
0.74
0.73
0.7
0.69
0.73
0.74
0.75
0.73
0.68
0.68
0.68
0.74
0.71
0.72
0.72
0.71
0.74
0.69
0.75

0.54
0.55
0.55
0.55
0.54
0.5
0.48
0.55
0.56
0.56
0.55
0.48
0.48
0.48
0.56
0.52
0.54
0.54
0.53
0.56
0.49
0.56

C. Acquired Customer Intimacy at the Individual Level

325

Table C.2.: Prediction of the Variable Knowledge High at the Individual

Level: Best Configurations and Results

Decision
Tree C4.5

Attribute

Value

Attribute

Value

BinarySplit
Minimum
Number of Object
per Leaf
Reduced Error
Pruning
Unpruned

False

Confidence Factor

0.4

3

Number of folds

N/A

False

SubTreeRaising

True

False

UseLaplace

False

Value

Performance
Indicator

Value

84.0
78.4

Recall(%)
F-Measure (%)

70.0
75.0

Attribute

Value

Performance
Indicator
Precision (%)
Success Rate (%)
Kappa Statistic
(%)
Attribute
k-nearest
Neighbor

56.0
Value

No
Distance
Weighting
LinearNNSearch

kNN

9

distanceWeighting

meanSquared

False

NearestNeighbor
SearchAlgorithm

Performance
Indicator

Value

Performance
Indicator

Value

83.0
77.1

Recall(%)
F-Measure (%)

67.0
72.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

54.0

326

Appendix

Prediction of the Variable Knowledge High at the Individual Level:
Best Configurations and Results (Continued)

Support
Vector
Machine

Attribute

Value

Attribute

Value

Complexity c

1.5

PolyKernel

Epsilon

1.0.E−12

Kernel
Tolerance
Parameter

Performance
Indicator

Value

Performance
Indicator

Value

87.0
79.1

Recall(%)
F-Measure (%)

67.0
74.0

Value

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

Neural
Network

0.001

58.0

Attribute

Value

Attribute

Autobuild
hiddenLayer

True
a

Momentum

0.17

Reset
Training Time

True
500

Decay
False
LearningRate
0.1
Nominalto
True
BinaryFilter
ValidationSetSize
0
ValidationThreshold 20

Performance
Indicator

Value

Performance
Indicator

Value

87.0
80.2

Recall(%)
F-Measure (%)

71.0
76.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

60.0

C. Acquired Customer Intimacy at the Individual Level

327

Table C.3.: Prediction of the Variable Knowledge Very High at the Indi-

vidual Level: Best Configurations and Results

Decision
Tree C4.5

Attribute

Value

Attribute

Value

BinarySplit
Minimum
Number of Object
per Leaf
Reduced Error
Pruning
Unpruned

False

Confidence Factor

0.4

6

Number of folds

N/A

False

SubTreeRaising

True

False

UseLaplace

False

Value

Performance
Indicator

Value

72.0
79.4

Recall(%)
F-Measure (%)

55.0
59.0

Attribute

Value

Performance
Indicator
Precision (%)
Success Rate (%)
Kappa Statistic
(%)
Attribute
k-nearest
Neighbor

46.0
Value

No
Distance
Weighting
LinearNNSearch

kNN

3

distanceWeighting

meanSquared

False

NearestNeighbor
SearchAlgorithm

Performance
Indicator

Value

Performance
Indicator

Value

83.0
83.5

Recall(%)
F-Measure (%)

57.0
64.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

55.0

328

Appendix

Prediction of the Variable Knowledge Very High at the Individual
Level: Best Configurations and Results (Continued)
Attribute
Support
Vector
Machine

Attribute

Value

Complexity c

1.7

Kernel

Normalized
PolyKernel

Epsilon

1.0.E−12

Tolerance
Parameter

0.001

Performance
Indicator

Value

Performance
Indicator

Value

71.0
80.3

Recall(%)
F-Measure (%)

62.0
63.0

Value

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

Neural
Network

Value

50.0

Attribute

Value

Attribute

Autobuild
hiddenLayer

True
a

Momentum

0.1

Reset
Training Time

True
100

Decay
False
LearningRate
0.2
Nominalto
True
BinaryFilter
ValidationSetSize
0
ValidationThreshold 20

Performance
Indicator

Value

Performance
Indicator

Value

61.0
76.5

Recall(%)
F-Measure (%)

60.0
58.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

42.0

C. Acquired Customer Intimacy at the Individual Level

329

Table C.4.: Prediction of the Variable Relationship High at the Individual

Level: Best Configurations and Results

Decision
Tree C4.5

Attribute

Value

Attribute

Value

BinarySplit
Minimum
Number of Object
per Leaf
Reduced Error
Pruning
Unpruned

False

Confidence Factor

0.20

2

Number of folds

N/A

False

SubTreeRaising

True

False

UseLaplace

False

Value

Performance
Indicator

Value

80.0
67.0

Recall(%)
F-Measure (%)

59.0
65.0

Attribute

Value

Performance
Indicator
Precision (%)
Success Rate (%)
Kappa Statistic
(%)
Attribute
k-nearest
Neighbor

35.0
Value

No
Distance
Weighting
LinearNNSearch

kNN

6

distanceWeighting

meanSquared

False

NearestNeighbor
SearchAlgorithm

Performance
Indicator

Value

Performance
Indicator

Value

80.0
73.3

Recall(%)
F-Measure (%)

75.0
76.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

45.0

330

Appendix

Prediction of the Variable Relationship High at the Individual Level:
Best Configurations and Results (Continued)

Support
Vector
Machine

Attribute

Value

Attribute

Value

Complexity c

1.2

PolyKernel

Epsilon

1.0.E−12

Kernel
Tolerance
Parameter

Performance
Indicator

Value

Performance
Indicator

Value

86.0
75.4

Recall(%)
F-Measure (%)

69.0
75.0

Value

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

Neural
Network

0.001

51.0

Attribute

Value

Attribute

Autobuild
hiddenLayer

True
a

Momentum

0.1

Reset
Training Time

True
50

Decay
False
LearningRate
0.2
Nominalto
True
BinaryFilter
ValidationSetSize
0
ValidationThreshold 20

Performance
Indicator

Value

Performance
Indicator

Value

79.0
70.9

Recall(%)
F-Measure (%)

72.0
73.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

41.0

C. Acquired Customer Intimacy at the Individual Level

331

Table C.5.: Prediction of the Variable Relationship Very High at the Indi-

vidual Level: Best Configurations and Results

Decision
Tree C4.5

Attribute

Value

Attribute

Value

BinarySplit
Minimum
Number of Object
per Leaf
Reduced Error
Pruning
Unpruned

False

Confidence Factor

0.1

7

Number of folds

N/A

False

SubTreeRaising

True

False

UseLaplace

False

Value

Performance
Indicator

Value

75.0
81.1

Recall(%)
F-Measure (%)

52.0
58.0

Attribute

Value

Performance
Indicator
Precision (%)
Success Rate (%)
Kappa Statistic
(%)
Attribute
k-nearest
Neighbor

48.0
Value

Weight
by 1 distance
LinearNNSearch

kNN

5

distanceWeighting

meanSquared

False

NearestNeighbor
SearchAlgorithm

Performance
Indicator

Value

Performance
Indicator

Value

66.0
77.8

Recall(%)
F-Measure (%)

55.0
57.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

43.0

332

Appendix

Prediction of the Variable Relationship Very High at the Individual
Level: Best Configurations and Results (Continued)

Support
Vector
Machine

Attribute

Value

Attribute

Value

Complexity c

1

PolyKernel

Epsilon

1.0.E−12

Kernel
Tolerance
Parameter

Performance
Indicator

Value

Performance
Indicator

Value

65.0
77.4

Recall(%)
F-Measure (%)

52.0
54.0

Value

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

Neural
Network

0.001

41.0

Attribute

Value

Attribute

Autobuild
hiddenLayer

True
a

Momentum

0.2

Reset
Training Time

True
2000

Decay
False
LearningRate
0.3
Nominalto
True
BinaryFilter
ValidationSetSize
0
ValidationThreshold 20

Performance
Indicator

Value

Performance
Indicator

Value

50.0
74.9

Recall(%)
F-Measure (%)

51.0
47.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

34.0

XC

h a n g e Vi
e

w

O
N
y
to

bu
lic

k

333

D. Acquired Customer Intimacy at the
Organizational Level

.d o

m

w

.c

o

c u -tr a c k

C

D. Acquired Customer Intimacy at the Organizational Level

m

.c

o

.d o

w

w

w

w

w

C

lic

k

to

bu

y

N

O

W

!

F-

er

W

w

PD

h a n g e Vi
e
!

XC

er

PD

F-

c u -tr a c k

GET DATA
/TYPE=XLS
/FILE='C:\Data\Service Research\Projects\CIG Project Local\Statistic Analysis Organization\110519\1105
/SHEET=name 'Acquired Organisation Result'
/CELLRANGE=full
/READNAMES=on
/ASSUMEDSTRWIDTH=32767.
EXECUTE.
DATASET NAME DataSet1 WINDOW=FRONT.
SAVE OUTFILE='C:\Data\Service Research\Projects\CIG Project Local\Statistic Analysis '+

This 'Organization\110519\110513_CI_Organisation_5_Knowledge.sav'
appendix complements chapter 7 and provides further details
/VERSION=2
on the
performed calibration to predict the acquired customer inti/COMPRESSED.
macy
components at theResearch\Projects\CIG
organizational Project
level. Local\Statistic
Figure D.1 Analysis
outlines
SAVE OUTFILE='C:\Data\Service
'+
'Organization\110519\110513_CI_Organisation_5_Knowledge.sav'
the Crombach’s
Alpha values of the summated scales Knowledge and
/COMPRESSED.
Relationship
RELIABILITY at the organizational level. Tables D.1, D.2, D.3 and D.4
/VARIABLES=Question11
Question13
present
the settings Question12
of four considered
machine learning algorithms
/SCALE('Knoledge Scale') ALL
which
achieved the best prediction of the variables Knowledge High,
/MODEL=ALPHA.
Knowledge Very High, Relationship High, and Relationship Very High at
Reliability
the
organizational level.
[DataSet1] C:\Data\Service Research\Projects\CIG Project Local\Statistic A
nalysis Organization\110519\110513_CI_Organisation_5_Knowledge.sav

Scale: Knoledge Scale

Scale: Relationship

Case Processing Summary
N
Cases

Valid

%
77

Excluded

a

Total

Case Processing Summary
N

100.0

0

.0

77

100.0

Cases

Valid
a

Excluded
Total

%
73

100.0

0

.0

73

100.0

a. Listwise deletion based on all
variables in the procedure.

a. Listwise deletion based on all
variables in the procedure.

Reliability Statistics

Reliability Statistics

Cronbach's
Alpha
.911

N of Items

Cronbach's
Alpha

N of Items

.891

3

3

Statistics
(b) ScaleItem
Relationship

(a) Scale Knowledge

FREQUENCIES VARIABLES=Knowledge
/ORDER=ANALYSIS.

Mean

Std. Deviation

N

4.74
Figure D.1.: Crombach’s Alpha of theQuestion14
Scales Knowledge
and1.608
Relation- 73
Frequencies
4.29
1.532
73
ship at the OrganizationalQuestion15
Level
Question16

4.64

1.584

73

[DataSet1] C:\Data\Service Research\Projects\CIG Project Local\Statistic A
nalysis Organization\110519\110513_CI_Organisation_5_Knowledge.sav
Scale Statistics
Mean
13.67

Variance

Std. Deviation

18.335

4.282

N of Items

Page 1

3

334

Appendix

Table D.1.: Prediction of the Variable Knowledge High at the Organiza-

tional Level: Best Configurations and Results

Decision
Tree C4.5

Attribute

Value

Attribute

Value

BinarySplit
Minimum
Number of Object
per Leaf
Reduced Error
Pruning
Unpruned

False

Confidence Factor

0.2

4

Number of folds

N/A

False

SubTreeRaising

True

False

UseLaplace

False

Value

Performance
Indicator

Value

73.0
79.6

Recall(%)
F-Measure (%)

65.0
66.0

Attribute

Value

Performance
Indicator
Precision (%)
Success Rate (%)
Kappa Statistic
(%)
Attribute
k-nearest
Neighbor

53.0
Value

No
Distance
Weighting
LinearNNSearch

kNN

10

distanceWeighting

meanSquared

False

NearestNeighbor
SearchAlgorithm

Performance
Indicator

Value

Performance
Indicator

Value

73.0
78.9

Recall(%)
F-Measure (%)

51.0
58.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

47.0

D. Acquired Customer Intimacy at the Organizational Level

335

Prediction of the Variable Knowledge High at the Organizational Level:
Best Configurations and Results (Continued)

Support
Vector
Machine

Attribute

Value

Attribute

Value

Complexity c

2.5

RBFKernel

Epsilon

1.0.E−12

Kernel
Tolerance
Parameter

Performance
Indicator

Value

Performance
Indicator

Value

75.0
82.1

Recall(%)
F-Measure (%)

81.0
75.0

Value

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

Neural
Network

0.001

61.0

Attribute

Value

Attribute

Autobuild
hiddenLayer

True
a

Momentum

0.2

Reset
Training Time

True
150

Decay
False
LearningRate
0.2
Nominalto
True
BinaryFilter
ValidationSetSize
0
ValidationThreshold 20

Performance
Indicator

Value

Performance
Indicator

Value

50.0
68.3

Recall(%)
F-Measure (%)

54.0
48.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

29.0

336

Appendix

Table D.2.: Prediction of the Variable Knowledge Very High at the Orga-

nizational Level: Best Configurations and Results

Decision
Tree C4.5

Attribute

Value

Attribute

Value

BinarySplit
Minimum
Number of Object
per Leaf
Reduced Error
Pruning
Unpruned

False

Confidence Factor

0.2

2

Number of folds

N/A

False

SubTreeRaising

True

False

UseLaplace

False

Value

Performance
Indicator

Value

41.0
84.6

Recall(%)
F-Measure (%)

45.0
40.0

Attribute

Value

Performance
Indicator
Precision (%)
Success Rate (%)
Kappa Statistic
(%)
Attribute
k-nearest
Neighbor

35.0
Value

No
Distance
Weighting
LinearNNSearch

kNN

4

distanceWeighting

meanSquared

False

NearestNeighbor
SearchAlgorithm

Performance
Indicator

Value

Performance
Indicator

Value

42.0
87.6

Recall(%)
F-Measure (%)

39.0
38.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

36.0

D. Acquired Customer Intimacy at the Organizational Level

337

Prediction of the Variable Knowledge Very High at the Organizational
Level: Best Configurations and Results (Continued)

Support
Vector
Machine

Attribute

Value

Attribute

Value

Complexity c

0.4

PolyKernel

Epsilon

1.0.E−12

Kernel
Tolerance
Parameter

Performance
Indicator

Value

Performance
Indicator

Value

32.0
80.1

Recall(%)
F-Measure (%)

35.0
32.0

Value

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

Neural
Network

0.001

23.0

Attribute

Value

Attribute

Autobuild
hiddenLayer

True
a

Momentum

0.2

Reset
Training Time

True
400

Decay
False
LearningRate
0.2
Nominalto
True
BinaryFilter
ValidationSetSize
0
ValidationThreshold 20

Performance
Indicator

Value

Performance
Indicator

Value

14.0
80.1

Recall(%)
F-Measure (%)

19.0
14.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

10.0

338

Appendix

Table D.3.: Prediction of the Variable Relationship High at the Organiza-

tional Level: Best Configurations and Results

Decision
Tree C4.5

Attribute

Value

Attribute

Value

BinarySplit
Minimum
Number of Object
per Leaf
Reduced Error
Pruning
Unpruned

False

Confidence Factor

0.6

9

Number of folds

N/A

False

SubTreeRaising

True

False

UseLaplace

False

Value

Performance
Indicator

Value

55.0
56.0

Recall(%)
F-Measure (%)

55.0
52.0

Attribute

Value

Performance
Indicator
Precision (%)
Success Rate (%)
Kappa Statistic
(%)
Attribute
k-nearest
Neighbor

11.0
Value

No
Distance
Weighting
LinearNNSearch

kNN

4

distanceWeighting

meanSquared

False

NearestNeighbor
SearchAlgorithm

Performance
Indicator

Value

Performance
Indicator

Value

53.0
56.5

Recall(%)
F-Measure (%)

61.0
55.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

13.0

D. Acquired Customer Intimacy at the Organizational Level

339

Prediction of the Variable Relationship High at the Organizational
Level: Best Configurations and Results (Continued)

Support
Vector
Machine

Attribute

Value

Attribute

Value

Complexity c

2.5

RBFKernel

Epsilon

1.0.E−12

Kernel
Tolerance
Parameter

Performance
Indicator

Value

Performance
Indicator

Value

64.0
66.8

Recall(%)
F-Measure (%)

74.0
66.0

Value

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

Neural
Network

0.001

33.0

Attribute

Value

Attribute

Autobuild
hiddenLayer

True
a

Momentum

0.2

Reset
Training Time

True
9

Decay
False
LearningRate
0.1
Nominalto
True
BinaryFilter
ValidationSetSize
0
ValidationThreshold 20

Performance
Indicator

Value

Performance
Indicator

Value

54.0
58.6

Recall(%)
F-Measure (%)

67.0
58.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

17.0

340

Appendix

Table D.4.: Prediction of the Variable Relationship Very High at the Or-

ganizational Level: Best Configurations and Results

Decision
Tree C4.5

Attribute

Value

Attribute

Value

BinarySplit
Minimum
Number of Object
per Leaf
Reduced Error
Pruning
Unpruned

False

Confidence Factor

0.3

0.3

Number of folds

N/A

False

SubTreeRaising

False

False

UseLaplace

False

Value

Performance
Indicator

Value

48.0
78.2

Recall(%)
F-Measure (%)

49.0
44.0

Attribute

Value

Performance
Indicator
Precision (%)
Success Rate (%)
Kappa Statistic
(%)
Attribute
k-nearest
Neighbor

33.0
Value

No
Distance
Learning
LinearNNSearch

kNN

4

distanceWeighting

meanSquared

False

NearestNeighbor
SearchAlgorithm

Performance
Indicator

Value

Performance
Indicator

Value

41.0
79.6

Recall(%)
F-Measure (%)

29.0
33.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

24.0

D. Acquired Customer Intimacy at the Organizational Level

341

Prediction of the Variable Relationship Very High at the Organizational
Level: Best Configurations and Results (Continued)

Support
Vector
Machine

Attribute

Value

Attribute

Value

Complexity c

0.4

PolyKernel

Epsilon

1.0.E−12

Kernel
Tolerance
Parameter

Performance
Indicator

Value

Performance
Indicator

Value

32.0
80.1

Recall(%)
F-Measure (%)

35.0
32.0

Value

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

Neural
Network

0.001

23.0

Attribute

Value

Attribute

Autobuild
hiddenLayer

True
a

Momentum

0.2

Reset
Training Time

True
200

Decay
False
LearningRate
0.2
Nominalto
True
BinaryFilter
ValidationSetSize
0
ValidationThreshold 20

Performance
Indicator

Value

Performance
Indicator

Value

18.0
77.6

Recall(%)
F-Measure (%)

16.0
16.0

Precision (%)
Success Rate (%)
Kappa Statistic
(%)

11.0

342

Appendix

E. CI Analytics Implementation
This appendix complements chapter 6 and provides further details
on the implementation of the software CI Analytics. Section E.1 develops the services which have been conceived and implemented to
calculate the acquired customer intimacy metrics. Section E.2 elaborates on the services designed to calculate the leveraged customer
intimacy metrics. Subsequently, section E.3 introduces the software
CI Graph which is the first prototype of the software CI Analytics
and which was realized together with Thomas Herzig. Finally, section E.4 presents the questionnaire used to performed the survey on
the business benefits of the software CI Analytics and details the survey participants profiles.

E.1. CI Services for Calculating the Acquired Customer
Intimacy Metrics
This appendix elaborates on the CI Services realized to calculate
the acquired customer intimacy metrics which are presented in section 6.2.4.
As detailed in table E.1, the services focusing on the individual level
of analysis return a graph representing the social network formed by
the employees of the provider and of the customer. These employees
are represented by nodes on the graph and the customer intimacy
metrics values are indicated by the weights of the graph edges. The
social network graph returned by these CI Services is presented in
the XML format DyNetML which has been specifically conceived for
the representation of social networks (Tsvetovat et al., 2004). The services calculate the customer intimacy metrics upon the data available
in the customer interaction time fact table and take the seven following parameters as input:
• CustomerOrgRef determines the customer organization for which
the customer intimacy metrics are calculated
• StartDate and EndDate determine the beginning and the end of
the considered time frame.

E. CI Analytics Implementation

343

• SegmentSize specifies the length of each segment in the time
period and determines, therefore, the precision of the analysis.
• InteractionDurationThreshold, InteractionQuantityThreshold and
WeightedInteractionQuantityThreshold allow to further calibrate
the calculation of the customer intimacy metrics, as detailed in
section 5.2.2.1
The services realized to calculate the acquired customer intimacy at
the organization level of analysis return the value of the customer
intimacy metric between a specific provider employee and the considered customer organization. Similarly to the services created to calculate the acquired customer intimacy at the individual level, these
services use the data available in the customer interaction time fact
table in order to calculate the metrics. In addition to the input parameters defined for the services performing the calculation at the
individual level, the services calculating the acquired customer intimacy at the organizational level also take the reference to a specific
provider employee as input parameter, as depicted in table E.1.

344

Appendix

Table E.1.: CI Services For Calculating the Acquired Customer Intimacy Metrics: Technical Details
CI Services at the Individual Level
Name

Input
Parameters

Volume Service, WVolume Service, Intensity Service, WIntensity Service, Frequency Service, Duration Service, NumberEpisodes Service, Mode Service
CustomerOrgRef (String), StartDate (Integer), EndDate (Integer), SegmentSize (Integer), InteractionDurationThreshold (Integer), InteractionQuantityThreshold (Integer),
WeightedInteractionQuantityThreshold (Integer)

Output Value

Social Network Graph (DyNetML Format)

Fact Table

Customer Interaction Time Fact Table

Description

Eight CI services provide the functionality to calculate the
acquired customer intimacy metric at the individual level.

CI Services at the Organizational Level
Name

Input
Parameters

Org Volume Service, Org WVolume Service, Org Intensity
Service, Org WIntensity Service, Org Frequency Service,
Org Duration Service, Org NumberEpisodes Service, Org
Mode Service
CustomerOrgRef (String), ProviderEmployeeRef (String),
StartDate (Integer), EndDate (Integer), SegmentSize (Integer), InteractionDurationThreshold (Integer), InteractionQuantityThreshold (Integer), WeightedInteractionQuantityThreshold

Output Value

MetricValue (Double)

Fact Table

Customer Interaction Time Fact Table

Description

Eight organizational CI services provide the functionality
to calculate the acquired customer intimacy metrics at the
organizational level.

E. CI Analytics Implementation

345

E.2. CI Services for Calculating the Leveraged Customer
Intimacy Metrics
This appendix elaborates on the CI Services realized to calculate the
seven leveraged customer intimacy metrics which are presented in
section 6.2.4. Table E.2 describes each of these seven services and
provides information on their inputs and outputs.

Table E.2.: CI Services for the Leveraged Customer Intimacy Metrics
Service Name
Customization Revenue Ratio Service
Component

Customization

Metric

Customization Revenue Ratio

Input
Parameters

CustomerOrgRef (String), StartDate (Integer), EndDate (Integer)

Output
Parameter

Ratio value (Double) comprised between 0 and 1

Fact Table

Customer Value Return

Description

This service provides the functionality to calculate the customer intimacy metric Customization Revenue Ratio to assess
the degree of customization of the solution provided to the
customer. To perform the calculation, this service only uses
the monetary revenues derived from business objects of
type Invoice Line Item and excludes non monetary returns
such as customer suggestions.

Customer Purchase Frequency Ratio Service
Component

Customer Loyalty

Metric

Customer Purchase Frequency Ratio

Input
Parameters

CustomerOrgRef(String), StartDate (Integer), EndDate (Integer)

Output
Parameter

Frequency value (Double) comprised between 0 and 1

Fact Table

Customer Value Return (only monetary revenues)

346

Description

Appendix

This service provides the functionality to calculate the customer intimacy metric Customer Purchase Frequency Ratio
which has been established as an indicator of the customer
loyalty.

CrossSelling Revenue Ratio Service
Component

Cross-Selling

Metric

Cross-Selling Revenue Ratio

Input
Parameters

CustomerOrgRef (String), StartDate (Integer), EndDate (Integer)

Output
Parameter

Ratio value (Double) comprised between 0 and 1

Fact Table

Customer Value Return

Description

This service enables the calculation of the customer intimacy metric Cross-Selling Revenue Ratio. Similarly to the
Customization Revenue Ratio Service, the CrossSelling Revenue Ratio Service only considers the monetary revenues
recorded in the Customer Value Return fact table. This
service identifies the source of the revenues such as product and service reference numbers that the customer purchased for the first time within the time period. It then calculates the Cross-Selling Revenue Ratio as the ratio between
the revenues generated from these sources and the total revenues generated in the considered time period with the
customer.

CrossSelling Diversity Ratio Service
Component

Cross-Selling

Metric

Cross-Selling Diversity Ratio

Input
Parameters

CustomerOrgRef(String), StartDate (Integer), EndDate (Integer)

Output
Parameter

Ratio value (Double) comprised between 0 and 1

Fact Table

Customer Value Return

Description

This service provides the functionality to calculate the customer intimacy metric Cross-Selling Diversity Ratio upon
the monetary revenues recorded in the fact table Customer
Value Return.

E. CI Analytics Implementation

347

Customer Participation Quantity Service
Component

Customer Participation

Metric

Customer Participation Quantity

Input
Parameters

CustomerOrgRef (String), StartDate (Integer), EndDate (Integer)

Output
Parameter

Participation Quantity (Double)

Fact Table

Customer Value Return (excluding monetary revenues)

Description

This service calculates the metric Customer Participation
Quantity upon the data available in the Customer Value
Return fact table as the total number of suggestions submitted by the customer in the considered time frame.

Customization Participation Ratio Service
Metric
Input
Parameters

Customer Participation Ratio
CustomerOrgRef(String), StartDate (Integer), EndDate (Integer)

Output
Parameter

Ratio value (Double) comprised between 0 and 1

Fact Table

Customer Value Return

Description

This service provides the functionality to calculate the customer intimacy metric Customer Participation Ratio. It accesses the data contained in the Customer Value Return
fact table, calculates the number of suggestions performed
by the customer during a certain time frame, and divides
this value by the revenues generated with this customer
during the same time frame.

Transaction Effectiveness Ratio Service
Component

Transaction Cost Reduction

Metric

Transaction Effectiveness Ratio

Input
Parameters
Output
Parameter

CustomerOrgRef (String), StartDate (Integer), EndDate (Integer)
Ratio value (Double) comprised between 0 and 1

348

Appendix

Fact Table

Customer Value Return

Description

This service provides the functionality to calculate the customer intimacy metric Transaction Effectiveness Ratio. It accesses the data contained in the three fact tables Customer
Value Return, Customer Activity Time and Customer Interaction Time. The total interaction time and activity time
that occurred within the considered time period are divided by the total revenues generated with the customer
during the same time period.

E.3. CI Graph: A First Prototype of CI Analytics
This section introduces the first prototype called CI Graph of the software CI Analytics. CI Graph has been conceived and implemented to
generate the data set on which the calibration of the customer intimacy metrics presented in chapter 7 has been performed.
While the software CI Analytics in its current version includes both
the acquired and leveraged customer intimacy metrics and adheres
to business intelligence application standards, the objective of the
application CI Graph was to prove the feasibility of the measurement
of the acquired customer intimacy metrics and of the representation
of these metrics by means of a social network graph. Therefore, the
software CI Graph provides the functionality to measure and visualize the eight customer interaction time based acquired customer intimacy metrics which are presented in chapter 5 at both the individual
and organizational levels. CI Graph is also capable of calculating the
centrality metrics developed in section 5.2.3.
Figure E.2 illustrates the architecture of the software CI Graph. This
architecture consists of multiple modules developed in the C# language. In order to access the data contained in the database of the
application CAS genesisWorld, the Data Access module of CI Graph
does not use an ETL process directly accessing the database. Instead,
CI Graph requests and receives the data through the CAS genesisWorld server, using an API of the CAS genesisWorld server. Thus,
the performance impact on CAS genesisWorld is significantly higher
with CI Graph than with CI Analytics.

E. CI Analytics Implementation

349

User

Presentation Layer
User Interface

Graph Interface

User and Graph Interface Event Handlers
Application Layer
CAS gW Retrieval
Algorithm Module

Graph Metrics
Module

Graph Algorithm
Module

Graph Structure
Module

Data Layer
Data Access Functionality

MS SQL
Server
Compact

Figure E.2.: CI Graph: Architecture Overview
In order to calculate the customer intimacy metrics, the CAS gWRetrieval module retrieves the required interaction data from CAS GenesisWorld, calculates upon predefined calibration parameters (time
period T, segment size d, interaction quantity threshold b and weighted interaction quantity threshold wb) the customer interaction time
and weighted customer interaction time for each segment in the considered time period. The acquired customer intimacy metrics are subsequently calculated upon this data by the functions implemented
in the Graph Metrics module and stored into a table contained in a
database. The data in this table has been used to perform the calibration of the customer intimacy metrics presented in chapter 7. In
order to represent the customer intimacy metrics in the form of a social network, the Graph Structure module uses the data contained in
the previously populated table and creates the graph representation
of the social network upon customer intimacy metric selected by the
user. The Graph Algorithm module can finally be used in order to
calculate the network centrality metrics upon this graph. Further details on the architecture and implementation of the software CI Graph
are available upon request from the author.

350

Appendix

Figure E.3 illustrates the calibration panel of the software CI Graph.
The user enters its credentials and the location of the CAS genesisWorld server in order to connect to CAS genesisWorld server and to
retrieve the required interaction data. The user subsequently enters
the name of a customer organization and a date used to specify the
considered time period: the calculation is performed for the year preceding the date entered in this panel. Finally, the user clicks on the
button StartAnalysis in order to begin the metric calculation process.
The graph panel of the software CI Graph allows a visualization of
the acquired customer intimacy by means of a social network, as depicted in figure E.4. In this diagram, the rectangles in the top row
represent the customer employees and those in the bottom row the
provider employees. The values of the acquired customer intimacy
metrics are indicated by the weights of the edges on the graph. The
interface provides the ability to select between the three time periods, namely 3 months, 12 months or all-time, as well as to select a
metric to be visualized on the edges and a layout for the network
representation. Clicking on the button CalculateGraphMetrics initiates the process of calculating the network centrality metrics at the
organizational level.

E. CI Analytics Implementation

Figure E.3.: CI Graph: Calibration Panel

351

352

Appendix

Figure E.4.: CI Graph: Visualization Panel

E. CI Analytics Implementation

353

E.4. Business Benefits Analysis
This section complements the business benefits survey developed in
section 6.3.2. Figures E.5, E.6 and E.7 represent the questionnaire
designed to assess the business benefits of the software CI Analytics.
Figure E.8 provides further details on the survey participants with
regard to their role in the organization and to their interactions with
customers.

354

Appendix

Interview 1/3

Thank you for participating in this interview!
It will take you less than 10 minutes to complete it.
Please return your answers by Friday September 3rd to
Thomas Herzig:
thomas.herzig@student.kit.edu
What do I have to do?
1. Read and understand the context
2. Answer questions by ticking the appropriate box.

What will happen with my answers?
Your answers help us evaluate the business benefits of our research and our prototype. All
answers and responses will be handled confidentially and anonymously at all times.

What is this interview about?
We are conducting research on an automatic analysis of interaction between companies
(from interaction data contained in an enterprise IT system). The results are used in a
customer relationship management application that shows a social network between
employees of a service provider and employees of their customers. The objective of this
prototype is to help the service provider employees answer questions like:
• “We are starting a new project with a team from CustomerXY, does someone from my
company already know them?”
• “Have we cultivated our relationship with the customer during the last months?”

The Relationship Network Overview (simplified example)
Department A

Department B

Management

Customer
employees
The bigger the
connection,
the more
relationship
those persons
established.

Department 1

Department 2

Office of the CEO

You and your
colleagues

Figure E.5.: CI Analytics: Business Benefits Questionnaire (1/3)

E. CI Analytics Implementation

355

Interview 2/3
Please provide some information about your activities
Question1: What is your role inside your company?

d
Services

Sales

Marketing

Development

Management Other

Question2: How many customer companies have you worked with during the last year?

d
Less than 3

Between 3 an 10

More than 10

Question3: How many employees from customers did you have contact with during the last year?

d
Less than 10

Between 10 and 50

More than 50

Question4: How much of your time did you spend working with customers during the last year?

d
Less than 20%

Between 20 and 50%

More than 50%

Now please consider customers with whom you have worked with during the past year
and imagine you had the relationship network overview presented above available
Question5: I would use this overview to identify colleagues who have knowledge about the customer
organization (strategy, processes, organization, behaviour, etc.)
Strongly agree

Agree

No opinion

Disagree

Strongly disagree

Question6: I would use this overview to identify colleagues who have established relationships with
the customer employees
Strongly agree

Agree

No opinion

Disagree

Strongly disagree

Question7: This relationship network overview would help us share knowledge about the customer
inside our organization
Strongly agree

Agree

No opinion

Disagree

Strongly disagree

Question8: This relationship network overview would help us coordinate our activities towards the
customer and to be seen as one team by the customer
Strongly agree

Agree

No opinion

Disagree

Strongly disagree

Question9: Analyzing the evolution of this relationship network overview over time would help us
monitor the relationship with the customer (e.g. “Have we cultivated the relationship with the customer
during the last months?”)
Strongly agree

Agree

No opinion

Disagree

Strongly disagree

Question10: Together with other indicators such as sales results, this information would help us
compare the performance achieved with different customers and would help us in our choice to invest
in one or the other customer.
Strongly agree

Agree

No opinion

Disagree

Strongly disagree

Figure E.6.: CI Analytics: Business Benefits Questionnaire (2/3)

356

Appendix

Interview 3/3
You are almost done, two last questions:
Question11: I think such a visualization would be useful in our company
Strongly agree

Agree

No opinion

Disagree

Strongly disagree

Question12: I would have privacy concerns if this type of information was made available in my
company
Yes, strong
concerns

Yes, I have
concerns

No opinion

No concerns

No, absolutely no
concerns

you have further feedback or comments? Which features would you like to see in such a tool?
FigureDo
E.7.:
Write
here... CI Analytics: Business Benefits Questionnaire (3/3)

YOU ARE DONE!! THANK YOU FOR YOUR HELP!!!

(a) Question 1

(b) Question 2,3,4

Figure E.8.: Business Benefits Survey: Participants Profiles

François Habryn
CUSTOMER INTIMACY ANALYTICS

The ability to capture customer needs and to tailor the provided solutions accordingly, also defined as customer intimacy, has become a significant success
factor in the B2B space – in particular for increasingly “servitizing” businesses.
However, many organizations struggle with measuring and proactively managing the degree of customer intimacy established with their customers. The work
presented in this book aims to remedy this issue by providing a solution to the
assessment and monitoring of the key aspects of a customer intimacy strategy.
It leverages business analytics and social network analysis technology in order to
provide an accurate, real-time, and easily implementable assessment of customer
intimacy, thereby effectively complementing existing customer relationship management systems.
This book proposes a solid, innovative and clearly written contribution that
should be of interest to all business and IT leaders facing the challenges of customer intimacy (Prof. Dr. Gerhard Satzger).
François Habryn is a senior research associate at the Karlsruhe
Service Research Institute. He gained several years of experience
in IT consulting with IBM and holds a Ph.D. in economics from
the Karlsruhe Institute of Technology. François Habryn graduated
from the University of Technology of Compiègne in France with a
Master’s degree in computer science and from the Ecole Supérieure de Commerce de Paris (ESCP-Europe) with a Master’s degree
in European business.

ISBN 978-3-86644-848-3

9 783866 448483