Can AI Help Fix A Broken AML System?

Can AI Help Fix A Broken AML System?

Criminals are not the only elements that can vex financial institutions when it comes to protecting against money laundering. False positives — instances of potential but uncertain fraud — can significantly gum up the works when it comes to anti-money laundering (AML) efforts, leading to friction and massive manual review that can eat up costs and cut into profits.

That provides an opportunity for artificial intelligence (AI), which is the case that was made during a recent PYMNTS podcast discussion between Karen Webster and Akli Adjaoute, founder, president and CEO of Brighterion. But for various reasons, financial institutions — even as they face increasing pressure from politicians and regulators to improve AML monitoring and defenses — don’t yet seem in a rush to employ AI for such tasks.


And that’s a big problem, according to Adjaoute.

“We all know the system is broken,” he said. “The system is not working.”

AI Disconnect

The PYMNTS AI Innovation Playbook demonstrates that there is a disconnect between perceived value and actual adoption when it comes to general issues of fraud in the payments world — a gap that underscores the points made by Adjaoute in his discussion with Webster.

Most financial institutions, or 63.6 percent, believe AI systems are effective in reducing fraud, yet only 5.5 percent have implemented one. Despite lower-perceived value in data mining (28.4 percent) and business rule management systems (17.6 percent), nearly all (92.5 percent) use data mining, and 65 percent use business rule management systems (BRMS).

Various reasons help explain those numbers — even in the face of large fines levied on financial institutions that run afoul of AML rules. The costs of implementing AI systems, the hassle of replacing legacy technology, and the elimination of jobs tied to all that can work to delay deployments of the best AML technology, Adjaoute said. He also said that often misguided perceptions of the data requirements for AI tend to slow down or even prevent deployments.

More specifically, he told Webster, the idea that data — the fuel for all AI — must be clean is a hindrance.

“You have to work with the data you have,” he said. “It’s what you have. You have to match the data to the world you are involved in.”

Adjaoute, who served in the French military, likened the AI data process to the employment of a tank.

“A tank was not designed to work on a freeway,” he said. “You take it where it’s needed.”

Data always will have typos and other mistakes, and it will contain the challenges of matching slightly different company monikers — a business may have different names depending on the country involved, along with other discrepancies. That said, getting the data into a workable AI system designed to protect against money laundering is much more important than making sure that data is perfectly clean, he said.

False Positives

One of the main reasons for that comes down to those false positives, those signals that are markers of suspicion but not certainty when it comes to money laundering and other types of fraud that financial institutions must always guard against. As Adjaoute told it, false positives waste time, and time is always money, especially when considering the manual review that is often needed to analyze a false positive and determine its true meaning.

“You can get 50 alerts for every one that makes sense,” he told Webster, talking about those false positives. “With AI, you can get two. Think about when you have thousands or millions of them. How many people do you have to hire to deal with it?”

AI doesn’t just cull the herd when it comes to false positives, however. It can spot subtle connections and patterns that are strong signs of money laundering and other types of fraud — and such technology can keep learning instead of being tethered to relatively solid AML rules. And that can provide benefits to FIs.

“The reputational damage you get from fraud is massive,” he said. “You can lose partners and lose customers. The damage is not just the money.”

The use of AI to fight against money laundering and other fraud clearly has a way to go. But educational efforts are underway to help bank executives move past the common misconceptions — and at the same time, there seems little chance that regulatory pressure about AML efforts will ease anytime soon.