After a year of testing, Equifax said, it has determined that artificial intelligence can improve significantly the chances of fighting check fraud.
The check services division of the Atlanta credit and consumer information company last fall installed artificial intelligence software from Neuristics Corp. to reduce the risk of authorizing fraudulent checks. Equifax evaluates the background of people and companies that present checks to its more than 30,000 retail clients.
Millions of transactions were used to construct a model that determines whether funds are available to honor a check.
Since it began testing the system, Equifax said, the dollar value of checks it approves has increased 17% and the dollar value of returned checks has declined 3%. Check collections have increased 40%, and merchants have reduced processing times 15%. Equifax declined to release specific figures for competitive reasons.
Artificial intelligence "allows us to make more intelligent decisions so the retailer can accept more checks and make more sales," said Jeffrey Carbiener, senior vice president and general manager of Equifax check services in St. Petersburg, Fla.
Merchants swipe checks through a reader, which captures customer data and sends it to Equifax. The Neuristics software examines the customers' check-writing history and patterns, age, and the check number. The software also factors in the retailer's industry and location.
This yields a calculation of how likely the check is to bounce. An approval or disapproval message usually comes back in less than a second, Mr. Carbiener said.
Equifax uses this service in its check guarantee program for setting approval parameters and acquiring checks that bounce.
Previously, Equifax had used an in-house system that evaluated each data piece separately. The Neuristics system integrates the data to give a more complete picture of the check writer, said R. Paul Gorman, chief technology officer of Neuristics in Hunt Valley, Md.
"We brought all those pieces together into a single score," he said.
