We have used Data Mining in business contexts to identify:

1. which individuals are likely to go bankrupt by looking at the way they use their credit cards,
2. who will be interested in buying certain products,
3. how valuable a particular customer is,
4. who is a good risk for an auto loan,
5. what actions should be taken to get someone to pay their bill,
6. what tax returns are likely to be fraudulent,
7. the probability that a particular credit card stolen.

The Citibank Experience
We have, for instance, used data mining to identify people who would be likely to go delinquent on their credit cards. We looked at the behavior of the cardholder - how much they spent each month and on what, how they paid their bills, how often they revolved (did not pay the full amount each month), and other behavioral information. We then used a set of techniques we developed for Department of Defense projects to predict how much money could be expected to be made or lost from any one account. We were able to identify for Citibank more than $200 Million of previously unidentified exposure with no increase in false alarms.

The IRS Experience
This experience was applied to the detection of fraud in electronic tax returns for the IRS. At the time we started the project (1994), the IRS was charged with modernizing. This included a charge to use modern techniques to reduce tax fraud. In 1994, professionals, such as H&R Block, performed electronic tax filing for individuals. It was suspected that many of these professionally prepared returns were fraudulent. The IRS chose to target this group of filers for scrutiny. We applied some of the techniques we developed for the Citibank problem to automatically detect fraud for the IRS. Initial implementation increased the detection rate of the IRS by 8000% with no increase in false alarms. The software we developed is now fully implemented by the IRS.

The Mastercard Experience
MasterCard applied our fraud detection algorithms to identify merchants engaged in credit card fraud. This story is similar to the IRS story. In this case, merchants were the targets. The particular data flows accessible to MasterCard gave MasterCard a particular advantage in identifying merchant fraud. They were able to provide added value to their member banks by providing merchant fraud detection. When the project was completed, MasterCard mounted a massive information campaign to advertise this added value.