The co-opérative Desjardins' Movement is the largest and most important banking institution in Québec, with 1,329 branches serving 4.2 million members and holding combined assets in excess of $80bn. Because it is a co-opérative, each member has voting rights and the branches are independent. The Québec network is divided between 10 independent Fédérations, with a Confédération head office located in Lévis. In common with most banks, however, the co-opérative is reducing the tellers' service, increasing the use of ATM and other IT methods, and rationalising its staff.

Although the bank's main business is in current accounts, loans and mortgages, it also provides a wide range of other financial products through its subsidiaries, such as life and property insurance, money transmission, etc. Because each branch is independent and can decide which of the bank's products and policies to adopt, the Confédération has to market its products to its own branches as well as to the members.

The bank has therefore used Clustan to develop a typology of its members, for two purposes, as follows:

• To retain the loyalty of its members by designing the best possible financial products to meet their needs.
• To capture more market share by identifying profitable services which satisfy members' needs and improve market penetration.

All member transactions are held on the Confédération's central computer, and it was therefore necessary to install Clustan on this computer. We first used the Clustan procedure Cluster to complete a hierarchical cluster analysis by Ward's method on a representative sample of 16 000 members, according to 16 variables chosen to reflect the characteristics of their financial transaction patterns. The variables included mixed types (continuous and multi-state) and some missing values. We therefore used a mixed-data similarity coefficient, suitably transformed to a distance metric, with imputation for the missing values. From this analysis 30 types of members were identified according to their transaction patterns.

We next used Classify, a single-pass identification model, to classify all of the bank's 4.2m Québec members. This procedure compared each member with the hierarchical classification obtained previously, by top-down branching through the tree. It typically found the cluster of best fit after about 10-12 comparisons with the tree. The analysis also provided the similarity between each member and all 30 clusters. This was useful to help the financial managers identify those members whose financial transactions were clearly of one type, or possibly a combination of two or more types.

Lastly, we estimated the profitability of each transaction cluster and of the customer accounts individually. This was of direct use for portfolio management by branch managers, and also provided valuable information for market segmentation and marketing. For example, where we identified members with large transaction volumes through one account with capital or a loan elsewhere, we could suggest a more economical consolidating approach. We have also been able to suggest better diversification of members' capital, utilising a $60,000 guarantee by the Québec government which is available to all of the bank's members.

The results were also useful for marketing, enabling the bank to focus on products which have the best financial performance; to reduce direct mailing costs and increase response rates by targeting product promotions at those customer types most likely to respond; and consequently, to achieve better branding and customer retention. In this way, the bank can retain and win the business of more profitable customers at lower costs.