Repeat steps 3 to 5 until you have reached the pre- specified maximum number of clusters.

6. Repeat steps 3 to 5 until you have reached the pre- specified maximum number of clusters.

• ClustanGraphics (Clustan) • DecisionWORKS Suite (Advanced Software

Note that some methods go all the way to a single

\pplications)

cluster of all items. To identify the solution you want, iden-

• SPSS (SPSS)

tify where you have obtained the desired number of clus- • Poly Analyst Cluster Engine (Megaputer) ters, and stop.

• Sokal code (see Hand, 1981) Applying the hierarchical method to our matrix

above with a goal of three balanced clusters, the initial There are also many free codes available from acade- solution is

mic sites. Do a Web search to find them.

P A R T II DECISION SUPPORT SYSTEMS'

EFFECTIVE CLUSTER ANALYSIS/ are in the bank's marketing efforts. The bank can focus on DATA MINING APPLICATIONS

products and services that have the best financial perfor- mance and target them to appropriate customers. This

Goulet and Wishart (1996) provide an excellent example reduces mailing and other contact costs. Response rates of how a bank was able to classify its customers to dramat-

have been improved by targeting product promotions ically improve their financial services. The Co-operative achieving better branding and customer retention. In fact, Desjardin's Movement is the largest banking institution in

Quebec (Canada). When this analysis was done, there more profitable customers are retained at lower costs. were 1,329 branches and 4.2 million members. The organi-

zation had combined assets in excess of (Canada) $80 bil-

FURTHER READING

lion. It was in the process of reducing teller services, For more details on cluster analysis, algorithms, and soft- increasing ATM use and other IT methods, and reducing

ware, see Aldenderfer and Blashfield (1984), Aronson and staff. The organization, in addition to banking, offered

Iyer (20.01), Goulet and Wishart (1996), Hand (1981), products and services that included life and property

Klein and Aronson (1991), Romesburg (1984), and Zupan insurance, and several others. Since each branch is inde-

(1982). Also, since Web sites change almost daily, we rec- pendent, the Confederation needed to market its products

ommend that you perform a Web search on cluster, cluster

to both its branches and its members. At the start of the analysis, and cluster methods. There are excellent acade- study, the bank executives realized that they needed a

mic sites, many of which include free computer codes. typology of its members not only to retain customer loy-

alty, but also to capture more market share by identifying profitable services to satisfy members' needs and improve

A BETTER SOLUTION TO THE EXAMPLE market penetration.

Though not balanced, the cluster solution (1,3,4}, (3,5,6), The bank performed a cluster analysis of a sample of

and {5} has a value of 72, better than the solutions 16,000 members. By doing so, it identified 16 variables that

described earlier.

reflected the characteristics of financial transaction pat- terns. Thirty member types were identified. Next, all 4.2

Cluster Analysis References

million members were classified with best fits of the 16

Aldenderfer, M.S., and R.K. Blashfield. (1984). Cluster measures, which were used to place them into one or more

Analysis. Thousand Oaks, CA: Sage. of the 30 member type clusters.

Aronson, J.E., and L.S. Iyer. (2001). "Cluster Analysis." In N o w financial managers and analysts can identify

Encyclopedia of Operations Research & Management

members whose financial transactions fall into one or

Science, Norwell, MA: Kluwer.

more of the 30 clusters. Given a particular member's clus- Goulet, M., and D. Wishart. (1996, June). "Classifying a ter, the profitability of each transaction cluster and indi-

Bank's Customers to Improve Their Financial vidual customer accounts can be measured. Each branch

Services." In Proceedings of the Conference of the

manager can view his or her customers as investments in a

Classification Society of North America (CSNA),

portfolio, and particular market segments for the branch's

Amherst, MA.

products and services are readily identified. Hand, D. (1981). Discrimination and Classification. New The results are impressive. The bank can identify

York: Wiley.

members with large transaction volumes in one account Klein, G., and J.E. Aronson. (1991). "Optimal Clustering: by matching them to their other loan or insurance

A Model and Method." Naval Research Logistics, Vol. accounts. The managers can then suggest more economi-

38, No. 1.

cal consolidations of members' investments and loans,

Romesburg, H. (1984). Cluster Analysis for Researchers.

thus leading to a higher level of customer satisfaction. Belmont, CA: Lifetime Learning. Additionally, managers can suggest better diversification

Zupan, J. (1982). Clustering of Large Data Sets. New of members' investments. But the most impressive results

York: Research Studies Press.