Banking Taps Machine Learning to Spot Money Laundering Schemes

In April, Danske Bank lowered its 2019 outlook as it deals with the fallout of a $223 billion money laundering scheme. Swedbank’s stockholders are nervous about U.S.-scrutiny of its money laundering policies as it, too, has been caught up in the probe, Reuters reports. And a regional Russian bank owned by a former U.S. congressman from North Carolina had its license revoked on April 5th for violations related to money laundering rules, according to American Banker.

This despite the fact that, for decades, governments around the world have pressured banks and financial services firms to serve as partners to law enforcement when it comes to preventing criminals and corrupt public officials from disguising money gained through illicit means as legitimate income — money laundering — and moving it into the financial system.

Today, there are signs that the U.S. government is taking the first steps to require that financial services firms adopt artificial intelligence (AI) and machine learning-based tools to help root out money laundering.

The evolution of rules and standards that banks and financial services firms must follow to remain in compliance with the anti-money laundering (AML) requirements enshrined in the Bank Secrecy Act and other legislation has followed a predictable pattern in the past.

Typically, large institutions with the capacity to invest in cutting-edge technology and processes will attempt to innovate, finding ways to make compliance more efficient and effective. If regulators see value in a new approach, they will give the firms license to experiment with it.

Eventually, though, as new methods are perfected, what was once seen as an innovative BSA compliance technique starts to be seen as a best practice — something every bank should be doing.

For example, more than two decades ago, large banks that were sensitive to the headline risks of being caught with dirty money on their books sometimes went to great lengths to identify the beneficial owners of companies they did business with. It was a sometimes tortuous process that involved peeling back layers of shell companies to identify the true owner of a company or an asset.

Eventually databases and processes were developed on a large enough scale that regulators decided identifying beneficial owners was no longer optional. In May of 2018, a new customer due diligence rule made it a requirement.

AI and Machine Learning Can Improve Anti-Money Laundering Efforts

One can see the same sort of development beginning in the use of AI and machine learning in the anti-money laundering space.

For banks, BSA compliance has always been an extremely data-intensive process, and knowing who owns the companies they do business with is only one part of the puzzle. Banks need to verify that individual customers are who they say they are. They also want to know who their customers do business with, whether they are related or otherwise connected to public officials or known criminals and terrorists, and whether they exhibit patterns of behavior consistent with illegal activity.

Doing this effectively requires assembling and analyzing vast amounts of information from a wide array of sources — news reports, financial statements, corporate filings, judicial records and much more.

In December, the major financial services regulatory agencies in the U.S., including the Board of Governors of the Federal Reserve System, the Federal Deposit Insurance Corporation, the Financial Crimes Enforcement Network (FinCEN), the National Credit Union Administration, and the Office of the Comptroller of the Currency, issued a joint statement recognizing that many banks are experimenting with new technology to help them process that data more effectively.

“Some banks are experimenting with artificial intelligence and digital identity technologies applicable to their BSA/AML compliance programs,” they wrote. “These innovations and technologies can strengthen BSA/AML compliance approaches, as well as enhance transaction monitoring systems. The Agencies welcome these types of innovative approaches to further efforts to protect the financial system against illicit financial activity. In addition, these types of innovative approaches can maximize utilization of banks’ BSA/AML compliance resources.”

At the same time, FDIC Chairman Jelena McWilliams signaled that using big data to assist in anti-money laundering compliance won’t always be the province of the financial services industry’s biggest players.

“New technology, such as artificial intelligence and machine learning, can provide better strategies for banks of all sizes to better manage money-laundering and terrorist-financing risks, while reducing the cost of compliance,” she said.

Big Gains in Transaction Monitoring

Given that criminals also benefit from advances in technology, it’s actually past time for financial services firms to begin incorporating AI and machine learning into their compliance processes, says Patrick Craig the financial crime technology lead for EY’s Europe, Middle East, India and Africa practice.

“The current AML approach is struggling to keep pace with modern money laundering activity,” he wrote. “There is a real opportunity for AI not only to drive efficiencies, but more importantly to identify new and creative ways to tackle money laundering.”

One of the key challenges for bank anti-money laundering programs is transaction monitoring — the process if combing through transaction data to identify potential suspicious activity.

Current automated systems produce a large volume of “false positives.” That is, they flag as suspicious transactions that are actually normal activity, which requires further investigation by bank staff. Unfortunately, by some estimates, up to 99 percent of transactions flagged by current systems are false positives.

“AI offers immediate opportunities to significantly reduce operational cost with no detriment to effectiveness by introducing machine learning techniques at different stages of the transaction monitoring process,” Craig writes. “AI is also being increasingly applied to customer due diligence and screening controls using natural language processing and text mining techniques.”

AI Can Help With Know Your Customer Rules

So-called Know Your Customer rules — requiring banks to do extensive and ongoing due diligence on customers — are another area in which AI shows huge promise for improving both efficiency and performance, he said.

“AI could bring increased breadth, scale and frequency to holistic KYC reviews in a way that better integrates ongoing screening and monitoring analysis. Risk and detection models would assess and learn from a richer set of inputs and produce outcomes in the context of both the customer’s profile and behavior. By leveraging AI’s dynamic learning capability coupled with skilled investigators, this model could be used to augment operations, provide quality control and even be used to train new resources.”

According to Scott Zoldi, chief analytics officer at Fair Isaac Corp., AI and machine learning techniques can provide a “superhuman” boost to financial services professionals working in the AML area. But first, you have to be able to explain it to them.

The systems that are being developed, he said, are able to establish connections from disparate data sources to create a clearer picture of clients’ activity. For example, algorithms can be developed to identify clients whose activity trips a number of different threshold levels simultaneously. For example, a client whose average transaction by dollar amount, number of deposits originating overseas, and velocity of money flow to “risky” countries are all above a bank’s 90th percentile, would be flagged for extra scrutiny.

In the past, any single one of those triggers might have been enough to send up a red flag, resulting in a flood of false positives. But a system empowered with machine learning techniques can narrow the focus considerably.

According to Zoldi, results are “based on the totality of data used to construct the machine learning model and is probabilistic.” It can also return specific reasons why an individual was flagged, giving investigators a head start in the effort to determine whether further action is required.

“Machine learning for AML can help with key compliance challenges like false positives, gaining new insight and understanding of customer behavior, and providing decision logic that is clear and explainable. Clearly, machine learning technology adds a “‘superhuman”‘ boost to the efficacy of AML efforts!” Zoldi wrote.

Warning: Algorithms Can Be Biased

However, turning over much of the hard work of detecting possible money laundering activity comes with its own set of risks. Writing in the Harvard Business Review in August 2018, Lisa Quest, Anthony Charrie, Lucas du Croo de Jongh, and Subas Roy warned that algorithms are only as good as the people who create them, and can have bias and blind spots built into them inadvertently.

They recommend a robust system of back testing results, to protect against a variety of possible problems. The idea is to uncover issues, such as patterns of false positives, that could lead to discrimination allegations; systematic failure to recognize specific activities as suspicious; and failure to keep up with new approaches criminals develop to move illegitimate funds into the financial system.

“To prevent this from happening, companies need to create and test a variety of scenarios of cascading events resulting from AI-driven tools used to track criminal activities,” they wrote. “To outsmart money launderers, for example, banks should conduct ‘war games’ with ex-prosecutors and investigators to discover how they would beat their system.”

In order to understand the algorithms — or the outputs of algorithms — you need to use a system with “explainable AI.” Trey Grainger, Chief Algorithms Officer at Lucidworks, explained that if the machine is offering a suggestion “you want to understand exactly what logic and data went into that suggestion.” Grainger believes that open AI will be a must for regulated industries.

The bottom line, according to McKinsey & Company’s Stuart Breslow, Mikael Hagstroem, Daniel Mikkelsen, and Kate Robu is that banks that invest in AI and machine learning technology for anti-money laundering compliance, with the proper safeguards in place, can reap huge benefits. The technology can drive compliance-error rates down below 5 percent, compared to an industry average of about 30 percent. Additionally, the vexing problem of false positives can be cut nearly in half.

They write, “Banks that invest strategically in these three areas, rather than tactically reacting to market and regulatory changes, can over time substantially reduce their risk exposure and capture other substantial benefits.”

Rob Garver is a DC-based financial services reporter.