1. 1 Using a Fuzzy Classification Query Language for Customer Relationship Management Nicolas Werro University of Fribourg Switzerland

2. 2 Contents 1. Motivation 2. Fuzzy Classification 3. Fuzzy Classification Query Language 4. Fuzzy Customer Classes 5. Conclusion & Outlook

3. 3 Motivation

4. 4 Motivation Fuzzy Classification Query Language (fCQL) for the analysis of customer relationships : Fuzzy Classification Query Language (fCQL) for the analysis of customer relationships :  Improve customer equity  Automate mass customization  Launch and verify marketing campaigns  Analyze customers’ evolution

5. 5 Motivation The fCQL toolkit is a combination of relational databases and fuzzy logic : The fCQL toolkit is a combination of relational databases and fuzzy logic :  Reduce the complexity of the customer data  Extract valuable hidden information  Enable the use of non-numerical values  Query on a linguistic level  No data migration needed

6. 6 Fuzzy Classification

7. 7 Context Model Extend the relational model by a context model : Extend the relational model by a context model :  To every attribute Aj defined by its domain D(Aj), we add a context K(Aj)  A context K(Aj) is a partition of D(Aj) into equivalence classes  Relational database schema with contexts R(A,K) where A=(A1,…,An) and K=(K1(A1),…,Kn(An))

8. 8 Classification Example Classification of customers based on two attributes : Classification of customers based on two attributes :  Turnover : [0,1000] with low turnover [0,499] high turnover [500,1000]  Payment behaviour : {in advance, on time, behind time, too late} with attractive payment behaviour {in advance, on time} non-attractive payment behaviour {behind time, too late}

9. 9 Linguistic Variable Turnover low turnover linguistic variable term domain context 0499 500 1000 [500.. 1000] [0.. 499] high turnover

10. 10 Sharp Classification in advance on time 1000 500 499 0 D(Payment behaviour) D(Turnover) behind time too late C1 Commit Customer C2 Improve Loyalty C3 Augment Turnover C4 Don’t Invest attractive payment behaviour non-attractive payment behaviour high turnover low turnover

11. 11 Fuzzy Classification 0 in advance on time 1000 500 499 0 D(Payment behaviour) D(Turnover)  high turnover  low turnover 1 0 1  attractive payment behaviour  non-attractive payment behaviour behind time too late 0.33 0.66 C1 Commit Customer C2 Improve Loyalty C3 Augment Turnover C4 Don’t Invest Davis C1 Commit Customer

12. 12 Aggregation of Fuzzy Sets The minimum operator The minimum operator The γ-operator The γ-operator Where x  X, 0  γ  1

13. 13 Fuzzy vs. Sharp Classification Characteristics of a fuzzy classification : Characteristics of a fuzzy classification :  The elements can be classified in several classes  Each element has one or several membership degrees indicating to what extend it belongs to the different classes  Disappearance of the classes’ sharp borders  Accurate information of the classified elements

14. 14 Fuzzy Classification Query Language

15. 15 From SQL to fCQL Querying the database with the Structured Query Language (SQL) : Querying the database with the Structured Query Language (SQL) : select list of attributes from relation wherecondition of selection Querying the database with the fuzzy Classification Query Language (fCQL) : Querying the database with the fuzzy Classification Query Language (fCQL) : classify objects from relation withcondition of classification

16. 16 Examples Perform a classification of all customers : Perform a classification of all customers : classify customer from cust_relation Evaluate customers with low turnover : Evaluate customers with low turnover : classify customer from cust_relation withTurnover is low turnover Classify customers in class : Classify customers in class : classify customer from cust_relation withclass is Commit Customer

17. 17 Fuzzy Customer Classes

18. 18 Customer Equity The customer equity suggests that we should treat the customers to their real value The customer equity suggests that we should treat the customers to their real value  A sharp classification cannot drive customer equity as every customer of a class is treated the same  A fuzzy classification can realize the customer equity. The membership degrees of a customer in the different classes represent the real value of the customer and therefore can determine the privileges this customer deserves

19. 19 Customer Equity in advance on time 1000 500 499 0 D(Payment behaviour) D(Turnover) behind time too late attractive payment behaviour non-attractive payment behaviour high turnover low turnover C1 C2 C4C3 Smith Brown Ford Miller

20. 20 Customer Equity 0 in advance on time 1000 500 499 0 C1 C2 C4C3 D(Payment behaviour) D(Turnover)  high  low 1 0 1  attractive  non attractive behind time too late 0.33 0.66 Smith Brown Ford Miller Smith: C1:100; C2:0; C3:0; C4:0 Brown: C1:35; C2:17; C3:32; C4:16 Ford: C1:16; C2:32; C3:17; C4:35 Miller: C1:0; C2:0; C3:0; C4:100

21. 21 Mass Customization Example: calculation of a customized discount Example: calculation of a customized discount  Discount rates can be associated with each fuzzy class : C1 : 10%, C2 : 5%, C3 : 3%, C4 : 0%  The individual discount of a customer can be calculated as the aggregation of the discount of the classes he belongs to, in proportion of his membership degrees in the classes

22. 22 Mass Customization - Smith: 1 * 10% + 0 * 5% + 0 * 3% + 0 * 0% = 10% - Brown: 0.35 * 10% + 0.17 * 5% + 0.32 * 3% + 0.16 * 0% = 5.3% - Ford: 0.16 * 10% + 0.32 * 5% + 0.17 * 3% + 0.35 * 0% = 3.7% - Miller: 0 * 10% + 0 * 5% + 0 * 3% + 1 * 0% = 0% in advance on time 1000 500 499 0 C1 10% C2 5% C4 0% C3 3% D(Payment behaviour) D(Turnover) behind time too late Smith Brown Ford Miller Smith: C1:100; C2:0; C3:0; C4:0 Brown: C1:35; C2:17; C3:32; C4:16 Ford: C1:16; C2:32; C3:17; C4:35 Miller: C1:0; C2:0; C3:0; C4:100

23. 23 Marketing Campaign Launching a marketing campaign can be very expensive Launching a marketing campaign can be very expensive  Select the most appropriate customers  Verify the impact of the campaign in order to improve the target group

24. 24 Marketing Campaign (2/2)  negative loyalty 10 6 5 1 1000 500 499 0 C1 Commit Customer C2 Improve Loyalty C4 Don’t Invest C3 Augment Turnover D(Loyalty) D(Turnover)  high turnover  low turnover 1 0 0 1  positive loyalty test group

25. 25 Customers’ Evolution With a fuzzy classification, there is the possibility of monitoring the customers through the classes With a fuzzy classification, there is the possibility of monitoring the customers through the classes By observing the moves of the customers in the classes we can either detect a customer getting better and reinforce this tendency, either detect a customer leaving the best class and try to stop this move By observing the moves of the customers in the classes we can either detect a customer getting better and reinforce this tendency, either detect a customer leaving the best class and try to stop this move

26. 26 Conclusion & Outlook

27. 27 Conclusion The fuzzy classification, by giving a more precise information of the classified elements, allows to: Avoid the gaps between the classes Avoid the gaps between the classes  No more inequities Treat the customers to their real value Treat the customers to their real value  No ejection of potentially good customers and better retention of the top customers Better determine a subset of customers for a special action Better determine a subset of customers for a special action  More efficient marketing campaign Monitor the customers’ evolution Monitor the customers’ evolution

28. 28 Outlook Adaptation of the marketing mix theory to take advantage of the fuzzy classes Adaptation of the marketing mix theory to take advantage of the fuzzy classes Comparison of the fuzzy classification approach with clustering methods Comparison of the fuzzy classification approach with clustering methods

29. 29 Questions ?