Banks Take the Hint From Regulators and Test AI to Spot Criminals

Artificial intelligence has transformed some parts the banking industry, but one critical area—watching out for financial crimes such as money laundering—has been late to the party. 

 

The hold up? Lenders fear AI might be too good. Some worry the technology will reveal suspicious accounts that slipped through legacy systems, unearthing weaknesses in monitoring and exposing them to regulatory scrutiny. 

 

Older technology relied on fixed parameters set by humans. But AI, which already powers customer chatbots and automated stock trades for banks, also can be deployed in monitoring software that can adapt with each fishy account it spots, enabling them to better keep pace with criminals’ tactics. 

“The banks were saying that ‘I need a signal—I need a sign from regulators that I’m not going to get dinged,’” said Gary Shiffman, chief executive of compliance technology company Giant Oak and an adjunct professor at Georgetown University’s Center for Security Studies. 

 

That signal came last month, when financial regulators said banks wouldn’t necessarily be punished for previously undetected red flags they discover while tinkering with anti-money-laundering compliance technology such as AI.  

 

Lenders such as U.S. Bancorp have come to recognize the benefits of working closely with regulators on AI implementation. As a result of the statement, more financial services companies are following suit, seeking AI guidance from agencies such as the U.S. Treasury Department’s anti-money-laundering division. 

 

The Dec. 3 statement laid bare the challenges of adopting new forms of technology in a heavily regulated industry—and particularly in banking, as lenders recently have paid heavy penalties for not keeping AML procedures up to snuff. 

 

U.S. Bancorp, one of the nation’s biggest regional banks, began using AI in its anti-money-laundering compliance division two years ago to comply with the terms of a 2015 enforcement order from the Office of the Comptroller of the Currency, which cited the company for having gaps in oversight. 

 

The Minneapolis-based bank worked with the OCC to develop new ways to spot suspicious transactions using internally developed machine- learning technology. It also added two AI specialists to its compliance division. 

 

The investment has given the bank’s compliance officers the tools they need to move beyond a simple, rules-based system—for instance, pulling up a list of cash transactions based on size—to one that improves with every suspicious payment it uncovers. 

 

The company now has the tools to run more advanced searches, flagging customers who look statistically different than others based on more indicators. Its algorithms also account for suspicious transactions that are referred to the bank by branches and law enforcement agencies. 

 

The OCC lifted its enforcement order against the company in December. U.S. Bancorp’s deferred prosecution order from the Justice Department, issued in February and also related to AML deficiencies, remains in effect. 

 

“The joint statement by regulators on AML innovation reaffirmed our efforts in continuing to explore machine learning and artificial intelligence in our transaction monitoring regime,” said Dana Ripley, a U.S. Bancorp spokesman. 

 

Although regulators adopted a stance of leniency in their statement, they also reminded banks they have a legal obligation to protect the financial system from abuse by illicit actors, suggesting that lenders have little room for error once the compliance monitoring technology is fully implemented. Regulators recommend that lenders validate their technology before flipping the switch. 

 

Among the biggest promises AI offers to financial compliance departments is the prospect of reducing the amount of time employees spend chasing down false leads—cases, for instance, where customer payments are halted because they trip internal alerts, proponents say. 

 

Singapore’s United Overseas Bank Ltd. , for instance, recently tested machine-learning software developed by a local startup, Tookitaki, yielding promising results: false flags on consumer accounts declined by 60%, and by half for corporate customers. 

 

“Once you improve the number of false alerts, you reduce unnecessary customer interactions because [you] don’t have to ask for more information,” said Victor Ngo, group head of compliance at UOB. 

 

Other foreign banks, including U.K.-based Standard Chartered PLC and HSBC Holdings PLC, also have announced partnerships with software providers and upstart firms. 

 

Technology providers are looking to capitalize on the green light U.S. regulators have given the industry, and expect spending on compliance technology by banks to increase in the year ahead. 

 

International Business Machines Corp. in 2017 launched its AI-powered Watson product to help banks spot criminals. The product sifts through unstructured data, including the text of wire transfers, and spots suspicious accounts based on patterns of transactions. 

 

IBM has conducted pilot programs for nearly a dozen banks during the past year, said Marc Andrews, vice president for financial crimes and conduct risk. Of those tests, three banks—two U.S. banks and the U.S. operations of a foreign-owned bank—have moved into production, he said. 

 

The joint statement from regulators could prompt more pilot programs. “We believe that this is going to drive significant growth and investment,” he said.