HSBC Holdings PLC is testing software from Fair Isaac Corp. of San Rafael, Calif., that it hopes will provide instant fraud-risk analysis for credit card transactions without hurting customer service.
George Lennox, the London company's senior manager of group credit and risk, said Strategy Science for Fraud Referral is being tested at HSBC Bank in the United Kingdom.
"Because we've got tens of millions of customers, we've got to make the best decisions we can very quickly," Mr. Lennox said in an interview Aug. 19. "You don't want to offend good customers, but you don't want to pay out on a criminal transaction."
Strategy Science operates with Falcon Fraud Manager, a program that HSBC uses worldwide. (Fair Isaac acquired that product when it purchased its developer, HNC Software Inc., in August of last year.)
Frank J. McKenna, Fair Isaac's director of business consulting, said the two programs work in real time to analyze up to 250 variables (including pattern of card use, time of day, and the likelihood the customer will close the account if HSBC questions a legitimate transaction that produces a "false positive" warning) to answer one question: "What is the net profitability of referring that transaction?"
The fraud referral program is better than using automated rules to flag transactions, according to Mr. Lennox.
Last month Fair Isaac said that HSBC was the first to sign on for the system, which was introduced at a conference in June. In addition, the bank is using Fair Isaac's consulting services to fine-tune the application.
This is HSBC's fourth project with the Strategy Science family of products and the first to focus on fraud, Mr. Lennox said. Earlier efforts involved scoring systems to set credit limits for cards and consumer loans. While it can take months or years to determine whether customers overspend their credit limits, he said, "with fraud [detection] you can see success quite quickly - the customer will tell you straight away whether you're right or wrong."
Intervening at the point of sale is an extreme step, and it presents the risk of losing the transaction or even the customer, Mr. Lennox said. The goal of the new program is to let the bank more precisely target fraudulent transactions before taking such a step.
"There is a hierarchy of concern. The vast majority of transactions we like. We will transact and pay those," Mr. Lennox said. At the next level, a doubtful transaction that is for a small amount or that does not otherwise have a high priority will generate a letter to the customer.
"Then you get to increasing levels of concern," he said, "where you need to telephone the customer."
At the highest level, HSBC may intervene at the point of sale by asking the merchant to request additional identification if the software signals "a very high probability" that a transaction is fraudulent, Mr. Lennox said.
Losing legitimate transactions or customers' closing accounts because of false positives becomes expensive, he said. "We've got to maximize the customers' legitimate transactions and minimize the criminal transactions. Rule optimization is absolutely key to us."
Mr. McKenna said that Fair Isaac conducted lab tests using historical data from a cross-section of issuers. Older, rule-based models referred 0.18% of the transactions for point of sale verification and generated 35 false positives for each actual fraud detected. Strategy Science for Fraud Referral reduced that ratio to 26 for each case of fraud. In a second test in which the software was given different criteria, the referral percentage was cut in half, the false positive ratio was 21 for every fraud detected, and fraud-related losses fell 9%.
Fair Isaac has also developed a system called Fraud Predictor to help card issuers identify the merchants most likely to be victims of fraud.
Fraud Predictor is available only in the United States, the largest market for credit cards, and Canada.
