Home / Banking Strategies / Cognitive computing’s smart new approach to regulation and money laundering

Cognitive computing’s smart new approach to regulation and money laundering

If regulation is by the book these days, it might as well be from the “Harry Potter” series: the lost volume where brave bankers defend the castle from fire-breathing dragons that grow stronger, fiercer and more adept by the day. Harry recites spells. The dragons spit back and recite Federal Reserve Board Regulation Z, 12 CFR Part 226. (That’s enough to scare even the Dark Lord Voldemort.)

That’s the challenge banks face with the ever-increasing crush of rules, including those that govern anti-money laundering (AML).  What’s worse: Even as the regs get more expensive to address, the bad guys get more sophisticated and learn new ways to launder without a trace.

Now: Imagine having your own Dumbledore  to wizard your way through the phalanx of regulatory and money laundering dragons. To that end, IBM’s Watson cognitive computing technology has enough firepower, if you will, to tackle regulatory challenges and cut compliance costs, all while outflanking the crooks.

But sometimes even a mighty wizard can use a hand. In November, IBM acquired Promontory Financial Group, a D.C.-based financial regulatory advisory firm. How does the partnership set the stage for new levels of AML performance? To learn more, BAI Banking Strategies talked with Alistair Rennie, general manager, solutions, analytics at IBM. Joining him was Promontory CEO-founder Gene Ludwig, who’s also a former Comptroller of the Currency.

BAI: How did this partnership come about?

Gene Ludwig: We understood the need for new cognitive technology tools because we’re so active in the risk and compliance area for banks. When a potential money laundering transaction takes place at a bank, an alert is generated. The bank uses human beings to review each alert to determine if it is potentially a real incident. If they determine it is, the bank files a report of suspicious activity to the government.

Banks use thousands of people to review alerts. However, on any given day—because one crook is especially devious or an expert reviewer at the bank is having a bad day—there will be human mistakes. Technology offers us the opportunity to put a floor under the level of mistakes it’s mechanical and can check all the alerts uniformly. That capability is very valuable.

Alistair Rennie: From IBM’s perspective, we do a tremendous amount of work with companies on a global basis and we look to the pressure points banks face to see where we might help. Not surprisingly, compliance came to the top of the list as a rapidly growing expense critical to the safety and operations of the financial institution itself.

Most approaches to handling those pressures added more and more people to handle the burden. But looking forward, that prospect was not financially sustainable. That’s when we realized cognitive computing could make a difference.

BAI: How so?

Rennie: Banks deal with highly structured data and most of the information involved with regulatory compliance is highly unstructured: texts, complaint data bases, recordings of data with widely varying voices. Before the advent of cognitive learning, we were unable to analyze such unstructured data. Now we have a machine that can evaluate it and put it into context. Then, the machine can interact with a professional to help the bank make a decision based on the evidence.

To be sure, we’re not looking to shift the regulatory burden to a machine. We’re trying to give the experts in the bank very clear evidence, insights and a recommendation. By getting all this information and learning from decisions, we can help experts make a better call: The machine interacts with the experts.

BAI: How does this help the bank?

Rennie: You can look at all the unstructured data that might suggest money laundering. Or, if you’re looking for conduct issues inside the bank—a significant issue in the retail space these days—you can start to read and understand the customer complaint data base. Or you might review electronic communications for past patterns of conduct.  

You can direct the computers to constantly read and scan the horizons for either changes in regulations themselves, or external things that change the way regulation is applied or interpreted, and see how that might relate to a control framework at the bank.

This approach is no longer about programming a computer in a linear way. This is about teaching a system how to create a hypothesis and how to make increasingly better recommendations and decisions.

BAI: So the ability to monitor and analyze unstructured data marked a big step forward for cognitive computing?

Rennie: Absolutely. Take anti-money laundering, for example. Banks today have systems based on 1980s and 1990s technology that flag transactions based on looking at structured data. The challenge posed by that approach is that better than 80 percent of alerts generated by those systems are false positives—transactions that upon further scrutiny don’t reveal suspicious activity.

Ludwig: In some systems 98 percent of the alerts are false positives. It’s terrible. Just imagine if you get two million transactions a day, which some large banks do. That’s an enormous burden to review. Even some of the better systems at banks generate 70 percent to 80 percent false positives. That’s huge.

BAI: So no matter the system, the level of false positive alerts is the extremely high.

Rennie: You get a very significant improvement in quality and accuracy when you start looking at additional unstructured data around those transactions such as the customer involved, adverse news on companies, and so on—and then you apply analytics to those systems.

BAI: What is the potential for Watson in anti-money laundering?

Rennie: Over time, we’d like to get to the point where Watson can assist in the initial draft of the regulatory filing. This will create a virtuous cycle—a better and clearer understanding about how decisions were made—and provide more unstructured data for further analysis. This is one area where the ability to review and analyze unstructured data will make a very big difference quite rapidly. When you have a machine with these attributes, you can begin to do some amazing things.


Want more Banking Strategies? Sign up for our free newsletter!

Robert Stowe England is a financial journalist who writes about retail and investment banking, financial markets and investing strategies. He is the author of five books, including “Black Box Casino: How Wall Street’s Risky Shadow Banking Crashed Global Finance.”