Merchants are fighting fraud with machine-learning systems, a self-updating alternative to rules-based fraud detection methods.

Like a seasoned detective who can pick out patterns in crime scenes easier than rookies, a machine-learning system finds fraudulent patterns the more it observes the ecosystem it's integrated with, says Jason Tan, CEO of Sift Science, a fraud detection provider.

Last week, Sift Science received an $18 million round of funding led by Star Capital. "Our mission is to make machine-learning fraud detection accessible to merchants of all sizes," Tan says.

Machine-learning systems analyze activity in real time, determining which transactions are normal for the system it monitors and which are suspicious. In comparison, within rules-based systems merchants have to set rules for each type of behavior they want to flag.

For example, the number of digits in an email address is a good indicator of fraud, says Tan. "Jason123@gmail.com is more likely fraud than jason.tan@gmail.com," he says. "A rules-based system might flag a transaction if the email address has three or more digits, but that doesn't mean there isn't fraud happening when there's two numbers in the email address."

Also, a transaction is more likely to be fraudulent if the person's first and last names are lowercase, Tan says. And if a particular user is updating their credit card information a lot over a week period there is a higher chance of fraud, since credit cards typically last several years without needing to be replaced, he says.

According to Tan, one out of five orders gets flagged for manual review when using a rules-based system. Of those that are flagged, less than 25% are actually fraudulent transactions, he says.

"We've seen [Sift Science] customers that are able to get a 7% false positive rate...which is well below the industry average," Tan says.

Banks and merchants today are juggling the rapid adoption of electronic payments and a surge in the amount of data they handle, says Loc Nguyen, chief marketing officer at Feedzai, a machine-learning provider. "With these two trends we see a significant importance in artificial intelligence and computer machine-learning to keep up with the expanding data mass."

The Central Payment Network of Portugal, where Feedzai launched in 2009, uses the company's machine-learning technology. Poland and Romania also use Feedzai.

And these countries have some of the lowest card fraud rates throughout the world, according to the European Central Bank's Report on Card Fraud from February 2014. The rates in these countries are between four and ten times lower than card fraud rates in Great Britain and France, according to the report.

In January, Feedzai opened an office in Silicon Valley. "We're entering the U.S. market and think it is particularly timely as this market adopts mobile and particularly EMV-chip technology like that used in Europe," says Nguyen.

Machine-learning systems are scalable, making them more appropriate for large companies, says Tom Donlea, director of risk services at Whitepages Pro, which provides fraud detection services to clients. "It's just not sustainable for these large enterprise businesses to create rules upon rulesÂ…especially as they move into new areas [internationally], add products and services and just grow in general," he says.

Whitepages Pro tested its software with several merchants and saw that it reduced manual reviews by 40% during the first 90-day cycle, says Donlea. During the second 90-day cycle, manual reviews were reduced by 20%, he says.

The employees who handle these reviews can be freed up to handle revenue-generating roles, he says.

Subscribe Now

Authoritative analysis and perspective for every segment of the payments industry

14-Day Free Trial

Authoritative analysis and perspective for every segment of the industry