Discovering Customer Journey Maps

using a Mixture of Markov Models

Matthieu Harbich - Presenter

Gaël Bernard - University of Lausanne

Dr. Pietro Berkes - Kudelski

Pr. Benoît Garbinato

Pr. Periklis Andritsos

XES - Standard
CAR DEALER

Challenge

E-M Markov models

Case study

Feature 1

Feature 2

Feature 3
…
Feature 4

Conclusion
1
Customer journey Customer Journeys: Capture the Customer Experience.
Customer journey map
Customer journeys are becoming increasingly complex
XES - Standard
• Numerous channels / devices
• Increasing number of interactions
Challenge

E-M Markov models

Case study
« Improving understanding of customers has been ranked one of the
Feature 1 most important research challenges in the coming year [1] »

Feature 2

Feature 3

Feature 4

Conclusion
2
[1] Research priorities 2014-2016. Tech. rep., Marketing Science Institute (2016), http://www.msi.org/uploads/articles/MSI_RP16-18.pdf
Customer journey CJMs: A customer-centric approach.
Customer journey map

XES - Standard Customer Journey Maps (CJMs) are used to understand, discuss,
or improve the main paths in the usage of a service
Challenge

E-M Markov models Mostly used as a design thinking tool / strategic tool
Case study

Feature 1
Getting information
Feature 2
Trying
Feature 3 Buying

Feature 4 Sharing experience

Conclusion
3
Customer journey Strategy vs. Reality.
Customer journey map

XES - Standard
« People don’t behave like robots, and no matter
Challenge how well we craft an experience, they will not
E-M Markov models perceive exactly as we anticipate or hope » [2]

Case study

Feature 1

Feature 2

Feature 3

Feature 4

Conclusion
[2] Richardson, A.: Series on customer journey: Using customer journey maps to improve customer experience, using customer
journey maps to improve customer experience, touchpoint bring the customer experience to life. Harb Bus Rev (2010)
4
Customer journey XES: Discovering CJMs from Event Logs.
Customer journey map
• When customers interact with a service, it (often) leaves traces
XES - Standard • We can leverage these traces to provide fact-based insights on customer journeys
• Inspired from Process Mining
Challenge
• INPUT: The XES standard is adequate to store CJMs [3]
E-M Markov models

Case study

Feature 1

Feature 2

Feature 3

Feature 4

Conclusion
[3] Bernard, G., & Andritsos, P. (2017). A Process Mining Based Model for Customer Journey Mapping. 5
Customer journey Challenge: Summarize the Logs using K Representative Journeys.

Customer journey map

XES - Standard

Challenge
1 Event logs 2 CJM with actual journeys 3 CJM with representative journeys
E-M Markov models (e.g., XES) A 4 A 1, 3, 4
1 B D A B 3 B
1
Case study 2 E C G A C C
3 B D A F D D
4 B D F 5
Feature 1 E E
2 5, 2
5 E C G C A F F
Feature 2 G G
time time
Feature 3

Feature 4

Conclusion
6
Customer journey Challenge: Summarize the Logs using K Representative Journeys.

Customer journey map

XES - Standard

Challenge
1 Event logs 2 CJM with actual journeys 3 CJM with representative journeys
E-M Markov models (e.g., XES) A 4 A 1, 3, 4
1 B D A B 3 B
1
Case study 2 E C G A C C
3 B D A F D D
4 B D F 5
Feature 1 E E
2 5, 2
5 E C G C A F F
Feature 2 G G
time time
Feature 3

Feature 4

Conclusion
6
Customer journey Challenge: Summarize the Logs using K Representative Journeys.

Customer journey map

XES - Standard

Challenge
1 Event logs 2 CJM with actual journeys 3 CJM with representative journeys
E-M Markov models (e.g., XES) A 4 A 1, 3, 4
1 B D A B 3 B
1
Case study 2 E C G A C C
3 B D A F D D
4 B D F 5
Feature 1 E E
2 5, 2
5 E C G C A F F
Feature 2 G G
time time
Feature 3

Feature 4

Conclusion
6
Customer journey Markov models: Modelling sequences.
Customer journey map
« In probability theory, a Markov model is a stochastic
XES - Standard model used to model randomly changing systems. It is
assumed that future states depend only on the current state,
Challenge
not on the events that occurred before it. » - Paul A. Gagniuc
E-M Markov models
Markov model
Case study
c d b c d a
Feature 1
a b c d
Feature 2 a a
Feature 3 b b
c c
Feature 4 d d
Conclusion First probability vector Transition matrix

Gagniuc, Paul A. (2017). Markov Chains: From Theory to Implementation and Experimentation. USA, NJ: John Wiley & Sons. pp. 1–256. ISBN 978-1-119-38755-8. 7
Customer journey Expectation-Maximization on a Mixture of Markov Models.
Customer journey map

XES - Standard Input Algorithm Output

Challenge
Expectation
E-M Markov models
K Markov models
Case study Learning the parameters
XES of the K Markov models
Feature 1
K
Feature 2

Feature 3
Maximization
Feature 4

Conclusion Optimizing the log-likelihood of the data

XES - Standard
An activity-based travel survey conducted in the Chicago metropolitan area
Challenge over a demographic representative sample of its population.

E-M Markov models

Case study

Feature 1 29,542 journeys - 2,381 are unique

Feature 2 “being at home” → “going to school” → “having a meal” → “being at home”

16 types of events
Feature 3
Average number of events per journey: 4.8
Feature 4

Conclusion
http://www.cmap.illinois.gov/data/transportation/travel-survey 9
Customer journey Case Study: Results.
Customer journey map

XES - Standard

Challenge

E-M Markov models

Case study

Feature 1

Feature 2

Feature 3

Feature 4

Conclusion
10
Customer journey Feature 1: Each Model is Responsible for each Journey.
Customer journey map

XES - Standard

Challenge

E-M Markov models

Case study

Feature 1

Feature 2

Feature 3

Feature 4 …

Conclusion
11
Customer journey Feature 1: Each Model is Responsible for each Journey.
Customer journey map

XES - Standard

Challenge

E-M Markov models

Case study

Feature 1

Feature 2

Feature 3

Feature 4 …

Conclusion
11
Customer journey Feature 1: Each Model is Responsible for each Journey.
Customer journey map

XES - Standard

Challenge

E-M Markov models

Case study

Feature 1

Feature 2

Feature 3

Feature 4 …

Conclusion
11
Customer journey Feature 1: Each Model is Responsible for each Journey.
Customer journey map

XES - Standard

Challenge

E-M Markov models

Case study

Feature 1

Feature 2

Feature 3

Feature 4

Conclusion
12
Customer journey Feature 1: Each Model is Responsible for each Journey.
Customer journey map

XES - Standard

Challenge

E-M Markov models

Case study

Feature 1

Feature 2

Feature 3

Feature 4

Conclusion
12
Customer journey Feature 2: Limited Number of Parameters.
Customer journey map
As input:
XES - Standard

Challenge
X
Feature engineering
X
Distance metric

E-M Markov models

Feature 1
Parameters set by a human can have an important impact on the
Feature 2
representative journeys. The E-M on a mixture of Markov models allows to
discover the representative journeys much more naturally.
Feature 3

Feature 4

Conclusion
13
Customer journey Feature 3: Computing the Next Most Probable Event.
Customer journey map
a b c d
XES - Standard
a
Challenge b

E-M Markov models c

Case study d

Feature 1

Feature 2

Feature 3

Feature 4

Conclusion
14
Customer journey Feature 4: Handling Shifting Behaviors - Soft Clustering.
Customer journey map

XES - Standard c

T0
Challenge

E-M Markov models Model 1

Case study

Feature 1
T0
Feature 2

Feature 3

Feature 4 Model 3 Model 2

Conclusion
15
Customer journey Feature 4: Handling Shifting Behaviors - Soft Clustering.
Customer journey map

XES - Standard c d

T0 T1
Challenge

E-M Markov models Model 1

Case study

Feature 1
T1
Feature 2

Feature 3

Feature 4 Model 3 Model 2

Conclusion
15
Customer journey Feature 4: Handling Shifting Behaviors - Soft Clustering.
Customer journey map

XES - Standard c d b

T0 T1 T2
Challenge

E-M Markov models Model 1

Case study

Feature 1
T2
Feature 2

Feature 3

Feature 4 Model 3 Model 2

Conclusion
15
Customer journey Feature 4: Handling Shifting Behaviors - Soft Clustering.
Customer journey map

XES - Standard c d b c

T0 T1 T2 T3
Challenge

E-M Markov models Model 1

Case study

Feature 1
T3
Feature 2

Feature 3

Feature 4 Model 3 Model 2

Conclusion
15
Customer journey

Customer journey map

XES - Standard Conclusion

E-M Markov models Kudelski application

Case study Probabilistic approach robustness

Research in progress!
Feature 1

Feature 2

Feature 3

Feature 4 Q&A

Conclusion