Banking Dive Article
For those tasked with combating money laundering, artificial intelligence (AI) and machine learning can do more to help banks reduce spending while stopping more crime. Read Tim Mueller's article in Banking Dive that looks at the potential of AI and machine learning in anti-money laundering (AML).
Pity the poor banker. No, really.
There are few more socially acceptable targets of public scorn in American life than the men and women who make their money handling other people's. Some of that contempt may be deserved, but in every bank the world over, there is one group of employees that truly deserves your empathy: those tasked with trying to solve the problem of money laundering.
It's difficult to get a handle on just how much money is made every year from drugs, terrorism, human trafficking, and all other manner of criminal activity. But at the high end, the estimates are staggering — some $2 trillion globally.
The investigators at the Financial Crimes Enforcement Network (FinCEN), a division of the Treasury Department, are responsible for stopping the flow of illegal cash. But in reality, the people on the front lines of the anti-money laundering fight work at the banks themselves.
As one congressman told me recently, the government has largely outsourced its AML policing to the banks. And unfortunately, the well-meaning people now responsible are working with fewer resources than those trying to cheat. Which means they could use any help available, including assistance from the biggest current buzzwords in business.
Artificial intelligence and machine learning may still be merely shiny objects in some industries. But in AML, the question isn't what they can do for banks and the financial industry in 10 years, it's what they can do for them now.
To understand how, it helps first to know why AML is such a difficult problem.
Fraud is the most common headache in finance, but one banks are already well-equipped to deal with. The question is binary — was the charge on that credit card legitimate or was it not — and, therefore, is that much easier to answer.
By comparison, AML is all haystack and no needle.
Criminals who launder money tend to have the resources to cover their tracks using shell companies, shadow buyers, and other elaborate tactics available to the highest bidder.
Consider that in the U.S. alone, AML spending within banks sits at about $23 billion per year. And yet, in the past decade, FinCEN has fined those same institutions another $23 billion for not for catching enough crooks in the act.
It isn't that bankers don't want to find a solution. Nor that they're not spending. It's just that a solution is so hard to come by. Which brings us back to the buzzwords.
Systems already exist to automate the generation of alerts, the first step in identifying whether a transaction might be suspicious. The problem is that the vast majority of these alerts are not productive, meaning they are ultimately deemed as normal activity.
Yet machine learning is showing massive potential to reduce these false alerts while identifying actually suspicious activity. In some cases, false positive rates as high as 90% have been reduced to 50%. That means investigators can spend less time manually reviewing the alerts. And it increases the chance that those alerts will be productive and passed on to FinCEN for proper investigation.
Humans will ultimately still make the judgment call, filing suspicious activity reports only when they deem necessary. Machines aren't taking over the critical work. They're merely providing incremental increases in effectiveness that, over time, will do more to reduce AML spending while also reducing the amount banks pay in AML fines and allowing experts to spend more time on work to catch criminals.
This added effectiveness has not gone unnoticed among the audience that is hungriest for help. Five federal agencies in the U.S., including FinCEN, issued a joint statement in December on the new high-tech applications to AML, saying they "welcome these innovative approaches to protect the financial system against illicit activity." That statement came a month after the Monetary Authority of Singapore puts its own stamp of approval on the use of AI in combating AML.
When regulators — normally the most conservative members in the ecosystem — get this excited about an opportunity, the technology offers more than false promises.
Bankers as a group may not always be the most sympathetic characters. But the people we've tasked with solving such a complicated, massive and costly global problem deserve great tools in addition to understanding.