Can this tech make banks better at spotting money launderers?

IBM wants banks to not just spot crime but also identify criminals.

And to do that, the tech giant is turning to its financial services unit, borrowing predictive analytic techniques originally built to determine, for instance, the likelihood that a customer will buy a mortgage.

Instead of relying on more traditional markers, such as demographic detail, address or occupation, customers will be segmented by financial behavior, said Marc Andrews, vice president of Watson Financial Services at IBM.

Customers involved in money laundering, for instance, are deemed more likely to transact or use an ATM at off-hour times of the day. The platform’s profiles to identify criminal behavior are informed by ex-regulators and former bank employees with experience in financial investigation, Andrews said, adding that customer information is constantly being updated.

With the dynamic segmentation platform, IBM put customers into eight different segments and found that for six of those segments, no alerts were escalated by Watson’s artificial intelligence. For two of those segments, however, alerts were elevated 20% and 24% of the time, respectively.

Working with different clients, IBM found for every 100 suspicious activity alerts generated, only a couple of alerts actually resulted in a suspicious activity report. “So they are dealing with 98% false positives,” Andrews said. “Maybe 15% to 30% of those 100 cases will get escalated and turned into a case.”

Total number of AML enforcement actions involving US banks in 2017 and 2018 to date.

Rather than hunting for the sliver of customers flagged as actual criminals, banks can reduce false positives for anti-money-laundering surveillance by looking at a smaller customer segment that mimics the criminal behavior profile and is upgraded to an elevated risk level, Andrews said.

“We’re able to look at the behavior and say that customers that behave like this, almost all of their alerts end up getting dismissed and never get escalated,” he added. “We can also say that the ones that behave like this almost always get escalated. What we’ll do is we will monitor over time changes in behavior. A customer may move from one segment to another if their behavior changes.”

Too many false positives in AML detection makes it harder to find true criminals, said Aaron Klein, policy director of the Brookings Institution’s Center on Regulation and Markets. “When you’re looking for a needle in a haystack, more hay is the least helpful thing.”

IBM took a step toward making this vision a reality when it began testing a dynamic segmentation platform last year that integrated into Watson Financial Services. The company copied the segmentation model from the lead-generation side of its financial services business.

For example, the company was able to identify what behaviors a customer would exhibit right before applying for a mortgage, such as less large-ticket discretionary spending, charges at Home Depot and Lowe's, and transactions at convenience stores and gas stations in new ZIP codes.

IBM has piloted the platform at three institutions, one of which will take it live before the year ends.

Marc Andrews, vice president of Watson Financial Services at IBM

Technology can push better suspicious activity reporting than regulation can, Klein said. “The incentives are not aligned properly among bank regulators to reduce false positives,” he said. “Financial institutions are graded on following procedures as opposed to accuracy.”

For banks that have been mostly reactive as opposed to proactive, the platform may also be a way to get out from heavy spending on AML teams and technology, said Al Pascual, senior vice president of research at Javelin Strategy & Research.

“These types of criminal events haven’t diminished even with increased investment in people and technology at banks,” Pascual said. “You have an artificial floor no matter how much money you throw at it.”

AML and counterterrorist-financing compliance is a major source of compliance costs — one estimate suggests that the aggregate costs for large, complex institutions could top $8 billion a year.

“Ultimately it depends if this tech proves itself out from a cost savings perspective,” said Jackson Mueller, an associate director at the Milken Institute's Center for Financial Markets. “If I can onboard that tech effectively, I would imagine it would start slowing down the increase in headcount.”

The platform’s success will also depend on its ability to mesh with the other technologies connected to Watson, said Tony Kaus, a senior analyst at Aite Group. “All of these technologies only work if they are a symphony that is orchestrated to work together with the goal of emulating human behavior,” he said.

For reprint and licensing requests for this article, click here.
Financial crimes Regtech AML Fintech Practice management software KYC Artificial intelligence Risk management IBM RegTech Conference
MORE FROM AMERICAN BANKER