8. Vera Miguéis
vera.migueis@fe.up.pt
Ana Camanho
João Falcão e Cunha
acamanho@fe.up.pt
jfcunha@fe.up.pt
+351-91-254 1104
9. A service system is a configuration of
technology and organizational networks
designed to deliver services that satisfy
the needs, wants, or aspirations of
customers.
Firms, as service systems, need, want
and aspire to survive, prosper, grow
(sometimes also making profits ),
relying on customers for that.
10.
11. How can we use SSME Research in
order to help the firm and its
customers?
We are still in the way of finding the
answers…and also the right questions!
12. This work proposes a new method for promotions design,
informed by product associations observed in homogeneous
groups of customers.
The method is based on clustering techniques to segment
customers, and decision trees to characterize the segments
profile.
This analysis is followed by the identification of the products
usually purchased together by customers from each segment.
This enables regular customization of promotions to specific
groups of customers, having in mind improved satisfaction of their
needs, wants, and aspirations.
13. • Research motivation
• Literature review
– Segmentation
– Market basket analysis
• Methodology
• Case study
– Contextual setting
– Data
– Segmentation results
– Market basket analysis results
– Customer centered strategies
• Conclusions and future research
Contents
Contents
Mo5va5on
Literature
Methodology
Case
Study
Conclusion
13
14. • Evolution of marketing efforts in retailing companies
Few concerns about consumers
Competitors
proliferation
Need to keep customers
Time
Product centered strategies
Lifestyle
changes
Need to satisfy customer needs
Customer centered strategies
Contents
Contents
Mo5va5on
Literature
Methodology
Case
Study
Conclusion
14
15. Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
15
16. Classification Clustering
Association Forecasting
Visualization Regression
[Ngai et al (2009)]
Sequence Discovery
Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
16
17. • Market segmentation [Smith (1956)]
– Segmentation criteria:
• Geographic (initially)
• Demographic
• Volume of sales
• Perceived value for customers
• Lifestyle
• Psycographic
• Customer behaviour – inferred from transaction records available in large
databases, or surveys [e.g. Kiang et al. (2006), Min and Han(2005), Helsen and
Green (1991), Liu and Shih(2005)]
– In particular: Recency (date of the last purchase), Frequency and Monetary
(“RFM” model, [Bult and Wansbeek (1995)])
– Techniques for segmenting customers: Data mining clustering
Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
17
18. • Market Basket Analysis
– Applied to large databases (transactional)
– Application domains:
• Banking [e.g. Peacock (1998)]
• Telecommunication [e.g. Klenettinen (1999)]
• Web analysis [e.g. Tan and Kumar (2002)]
• Retailing [e.g. Chen et al. (2004)]
– Objectives:
• Cross-sales [e.g. Poel et al. (2004)]
• Product assortment [e.g. Brijs et al. (2004)]
Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
18
19. Customers segmentation K-means algorithm
Characterization of customers’ profile Decision tree
Market basket analysis (*) Apriori algorithm
(Agrawal and Srikant, 1994)
Design of customized promotions
Improvement of service levels
(*) market basket analysis within segments is very rare in the literature
Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
19
20. 14th February: Valentine’s Day ...
Enjoy Fine French Cuisine Alongside
Classic Opera with a Starter and Main
Course for Two People, plus a Glass of
Prossecco each at
Le Bel Canto Restaurant
20 / 29
21. • Chain of hypermarkets, supermarkets and small supermarkets;
• Two loyalty cards: approximately 80% of the purchases are done using
such cards.
• Two ways of segmentation:
– “Frequency and Monetary value” segmentation;
– Lifestyle segmentation;
• Customer segments are not used to differentiate customers in strategic
policies to promote loyalty:
– Discounts for specific products advertised in the store shelves and leaflets,
that are applicable to all customer with a loyalty card;
– Discounts on purchases done on selected days (percentual discount or
absolute discount on total value of purchases). These are applicable to
customers that present at the cash-point the discount coupon sent by mail;
– Discounts for specific products on selected days.
Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
Case
21
22. • Data available:
– Transactions for the last trimester of 2009
– Demographic information for each customer: residence postcode, city,
date of birth, gender, number of persons in the household
• Data analysed:
– Customers whose average amount of money spent per purchase was
up to 500€
– Customers whose average number of purchases per month is up to the
mean plus three standard deviations (11.7 visits per month)
» 2.142.439 customers
» 16.341.068 shopping baskets
Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
Case
22
23. • Segmentation variables:
– Average number of purchases made per month
– Average amount of money spent per purchase
• 5 clusters defined according to DB index and elbow curve
1.2
0.54
Davies Bouldin Elbow Curve
1
0.538
SumOfSquares/k
0.536
DB index
0.8
0.534
0.532 0.6
0.53
0.528 0.4
0.526
0.2
0.524
0.522 0
-1 1 3 5 7 9 11 0 2 4 6 8 10 12
Number of clusters (k) Number of clusters (k)
Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
Case
23
24. #Customers (%)
37%
27%
20%
8%
8%
Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
Case
24
25. • Clusters’ profile:
Avg.#
purchases
per
month
≤3.2 >3.2 >6.2
Avg.
Amount
money
spent
per
purchase
≤135.9 >135.9
Avg.#
purchases
per
month
≤1.5 >1.5
Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
Case
25
26. Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
Case
26
27. • Transactions were aggregated by customer
• The products were aggregated by subcategory
– Examples of rules obtained:
Cluster 4
Antecedent
Consequent
Hair Conditioner Shampoo
Tomatoes Vegetables for salad
Sliced ham Flemish cheese
Cabbage Vegetables for soup
Pears Apples
Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
Case
27
28. • Customer development:
– The company may issue a discount voucher at the PoS that
advertises a consequent product of the association rule, which
was not recently bought by the customer who bought the
corresponding antecedent product.
• Examples:
– In Cluster 4:
» Discount shampoo to customers that have bought
conditioner but did not buy shampoo.
» Discount vegetables for salad to customers that have
bought tomatoes but did not buy vegetables for salad.
Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
Case
28
29. This work proposes a new method for promotions design,
informed by product associations observed in homogeneous
groups of customers.
The method is based on clustering techniques to segment
customers, and decision trees to characterize the segments
profile.
This analysis is followed by the identification of the products
usually purchased together by customers from each segment.
This enables regular customization of promotions to specific
groups of customers, aiming at improved satisfaction of their
needs, wants, and aspirations.
30. • Data mining allows to find natural clusters of clients on large
retailing databases, by means of customer behaviour segmentation.
• Decision trees enable discovering the rules characterizing customer
segments.
• Market basket analysis within segments seems to show good
potential to support the design of customized promotions and
consequently the provision of better service to customers.
• In the future, we intend to interview panel customers belonging to
each cluster, in order to see if they consider that the service levels
are improving or can be improved.
• We also intend to monitor the evolution of the results of the
satisfaction surveys.
Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
Case
Conclusion
30
31. • What are the adequate promotions to improve service
levels?
• Are derived association rules more relevant than
creativity to design promotions?
• What “level” of segmentation should be used? No
segmentation? The one proposed here? Individual
segmentation?
• How important is it to listen to customers, in each
segment, and individually?
• …?
Contents
Contents
Mo5va5on
Literature
Literature
Methodology
Case
Study
Conclusion
Case
Conclusion
31