The management and maintenance of an operational KADS system development

Abstract

The Barclaycard Fraudwatch knowledge-based system identifies potentially fraudulent transactions on Visa and Mastercard credit cards. It makes use of two generic tasks, Select and Assess. Select reduces the number of transactions to be considered from about one million to about fifty thousand. Assess further processes these fifty thousand to arrive at a set of about four hundred accounts, which are output to fraud operations staff for further action. Of these four hundred, about fifteen will eventually turn out to be fraudulent.

Fraudwatch was developed by Touche Ross and Barclays, using the KADS method for knowledge-based system development. It is fully embedded within the Barclaycard suite of business and cardholder programs running on IBM mainframes. Select is largely written in COBOL, Assess is written in KnowledgeTool^™, and the peripherals and communications are written in JCL. Fraudwatch runs daily in batch mode. It saves about thirty percent of pre-status fraud losses on the products to which it is applied.

An initial implementation went live in November 1990. The system has undergone two cycles of knowledge maintenance, in order to adapt performance to organisational changes and changes in business objectives, and to enhance overall performance. Further systems and enhancements are planned during 1992 and 1993, which are intended to deploy Fraudwatch at earlier stages of transaction processing and to look at devolving processing.