As fraudsters gain experience, so can anti-fraud tools

Register now

The British machine learning technology company Featurespace is boosting its expansion strategy in the U.S. by arguing that fraud-fighting technology can improve on its own through experience.

While it's not necessarily that simple, security risk models can self-replicate to a degree that the technology does most of the heavy lifting, according to Martina King, CEO of Featurespace, a Cambridge, U.K.-based company that just opened offices in New York and Charlotte.

"You can teach the machine to 'learn' how to spot a high risk attack, and that can be retained into the model so it doesn't have to be written," said King.

Machine learning and artificial intelligence are often used interchangeably, and refer to programming that enables software to update and respond based on its own embedded analysis of new data and external events. It's relatively old technology, but it's finding traction at financial institutions and other companies that move money, given the pressure to quickly adapt to mobile commerce, the migration to faster payment processing and embedding payments into wearables and other connected devices.

These trends require faster action and adjustments to underlying systems than a developer software upgrade can provide, even one delivered remotely. That's leading to lots of deals between payment gateways and artificial intelligence and machine learning companies. PayU, for example, is applying machine learning to enable quick decisions on store credit at the point of sale.

Featurespace's flagship product, ARIC, monitors customer data and credit lifecycles in search of anomalies. It uses what King calls an adaptive algorithm that doesn't have to be reprogrammed or tuned. The technology was developed by Featurespace's team, which has its roots in Cambridge University research from Featurespace founder Dave Excell and Bill Fitzgerald, a professor at the university. Their work focused on the extraction of "meaning" from a large data set.

The technology connects with other fraud tools and data sources to feed this analysis, which Featurespace called adaptive behavioral analytics. "The learning happens in real time, so a response can be generated in real time," King said.

And machine learning is also gaining buzz as a way to combat payments fraud, which should give opportunity for Featurespace but also create competition with companies such as Signifyd.

Featurespace has picked up two clients in the U.S. thus far: The processor Total System Services and an unnamed U.S.-based bank. For TSYS, the Featurespace deployment is part of a broader strategy to bring speed to the fraud fight; the processor has also partnered with fraud prevention company Ethoca to mitigate chargebacks.

In Europe, Featurespace has worked with card issuers, gaming companies, lotteries and Vocalink Zapp, which uses Featurespace to combat mobile payments fraud for Zapp partners Barclays, Nationwide, Metro Bank and others.

"A fraud system for card-not-present should be able to identify previously unseen fraud attack types and determine between good and bad behavior and make an instant decision and have an instant response," King said.

With every iteration of improved detection technology, fraud goes down for a while until fraudsters figure out how to get around it, said Rick Oglesby, founder and president of AZ Payments Group.

"The machine learning concept is a good one because it enables detection methods to continuously adapt to changes in fraud methodologies," Oglesby said, adding it’s also a big step forward vs. rule-based detection solutions that fraudsters can figure out over time. "Machine learning logic is less transparent and evolving, so it’s harder for fraudsters to defeat."

For reprint and licensing requests for this article, click here.