NetGuardians has changed the game when it comes to spotting money laundering, by training machine-learning models to spot more cases – with up to 10 times fewer false alerts than the rules and parameter approaches currently adopted by banks, writes Jérôme Kehrli.
At the end of last year, the International Monetary Fund (IMF) urged policy-makers to do more to curb financial crime and money laundering. It cited the personal and emotional cost to those affected, lower revenues, higher costs, lost banking services, the fact that it can fund terrorism, and increased financial instability as motivation.
But policy-makers are already piling the pressure on banks to do more. Regulations and enormous fines have pushed the need to combat money laundering right up the banking agenda. Just look at how last year across Europe, the Middle East and Africa banks spent almost twice what they had the year before on compliance – $85 billion, according to research by Forrester Consulting. Yet still the criminals continue to successfully launder huge sums – somewhere between 2 and 5 percent of the world’s gross domestic product, or between €715 billion and €1.87 trillion.
Clearly, stopping financial crime on this scale is not about policy, but about efficacy. Particularly when it comes to laundering the proceeds of petty crime, which we understand is under sharper focus from the authorities than ever before.
The shortcomings of rules and parameters
The fact that any criminal – petty or professional – is still able to clean their ill-gotten cash through the banking system stems from the central role rules and their parameters play in banks’ anti-money-laundering efforts. As criminals adapt and evolve their behavior to evade existing rules, banks are forced to adopt more rules and add more parameters to those rules to make them more effective.
Here’s a very simple example to illustrate the point. A rule stipulates that any transaction into a bank account over a set amount must be investigated, looking at the counterparty and why it is being paid at the very least. If the bank is not satisfied by the answers given by the account holder, it must report the transaction to the authorities. But that rule will have qualifying parameters – for an individual, the limit might be $10,000, for a company it might be $100,000 and for a high-net-worth individual it could be $500,000. Indeed, for every customer of the bank, or for different peer groups within it, there are different parameters.
And a bank’s rules cover more than just a ceiling sum – country of origination, counterparty, timing, destination for an outgoing transaction, how old the account is and more. And each rule will have parameters – sometimes up to five or six. Add in the fact that those parameters need to be maintained according to the account’s usual activity, as well as other concerns that might take account of changing geopolitical circumstances, and the result is an overwhelming complexity that eats resources and still fails to catch the money-launderers.
The abject failure of the current approach is down to the binary nature of rules and their parameters. A transaction either breaks a particular rule and parameters, or it doesn’t. It cannot spot grey areas – where the criminals are skilled at operating.
 |
You may be interested in watching the recording of our latest webinar on Artificial Intelligence in Anti-Money Laundering: A New Era of Banking Compliance. Click here to watch on demand: https://info.netguardians.ch/en/form/webinar/ai-in-aml-navigating-the-new-frontier-in-banking-compliance |
A more sophisticated, effective approach
This led us to develop software that doesn’t rely on parameters, but instead uses machine learning to develop models that can accurately interpret activity in these grey areas. This enables banks to get rid of parameters and so simplify the maintenance of their analytics systems. But simplification is not enough on its own. As the IMF has pointed out, we need to spot and stop more money laundering – and catch the criminals. Our software accurately identifies the money laundering, reducing the number of investigations that must be carried out. Vitally, in testing, it also spotted more genuine incidents that need to be reported to the regulators.
We trained these models on historic data – situations such as alerts reported to the regulator by banks, alerts investigated by the bank but found to be innocent and not reported and, crucially, transactions the banks didn’t spot but which were subsequently flagged by the regulator. We used hundreds of different signals to build the model, enabling it to interpret the grey areas – for example, an unusual amount of money flowing in and out of the account, an unusual number of new payees or a transaction first below a threshold, then above it. This works so well that we saw up substantially fewer false alerts while all the known money-laundering transactions were spotted and some that previously slipped through were detected.
When combined with a bank’s rules, the results are game-changing. Our software offers banks an affordable way to monitor money laundering far more effectively, reducing their costs and freeing up resources for more value-added roles. But equally importantly, it offers a real opportunity to curb money laundering – both by professional criminal gangs and also petty criminals.
Given the priority that the authorities are placing on AML activities, the complexity and cost of the solutions currently employed by banks, and the poor record of success, it makes sense for banks to move off the back foot and onto the front by talking to us to find out more about our results.
Jérôme Kehrli is chief technology officer and financial crime fighter at NetGuardians
