1 A Classification Problem in the Credit Card Industry: A Comparison of Three Solutions Christopher H. Jolly GE Corporate Research and Development, Schenectady, NY 12301, USA Mukkai S. Krishnamoorthy Department of Computer Science, Rensselaer Polytechnic Institute, Troy, NY 12180, USA Abstract Classifying accounts is a critical part of efficiently servicing credit cards. A problem we call the Rule Coverage Problem (RCP) arrises when using rules for specifying a classification strategy, and we develop and compare three approaches for solving this problem. 1 Introduction and RCP GE Capital Services (GECS) is the world’s largest supplier and manager of private label credit cards, serving 60 mil- lion accounts for 300 retailers. With such a large number of accounts it is critical to perform all servicing functions effectively and efficiently since small improvements on a per-account basis have a significant impact on profits. Two major servicing functions are collections and marketing. Collecting from delinquent accounts consists of creating and implementing a collection strategy that allocates resources (e.g., collector phone calls and computer-generated letters) to delinquent accounts. For example, “good” customers should be separated from “bad” customers because we collect efficiently from them in different ways. An effective collection strategy maximizes payments less expenses over the long term while maximizing customer good will. A collection strategy is created either by intuition and experience or more recently by optimization programs such as PAYMENT [Makuch90]. Marketing active accounts consists of creating and implementing a marketing strategy that augment billing state- ments with promotional messages and inserts. For example, credit card insurance only needs to be marketed to those card holders who do not already have it, and a new store opening only needs to be marketed to those living (or work- ing or shopping) within a reasonable driving distance. An effective marketing strategy increases credit card usage, lowers advertising costs, and increases (credit card and retailer) brand loyalty. Thus, GECS needs to be able to classify accounts into homogeneous groups as shown in Figure 1. Intelligent System Large, Dynamic Account Base Classified Account Base Classification Needs Rules Figure 1: Classification Problem