Powering Anti-Money Laundering efforts with Artificial Intelligence

The proliferation of more powerful hardware is making it possible to apply increasingly advanced machine learning and artificial intelligence to assist with solving everyday problems, from face recognition on mobile phones to self-driving cars. One area that is seeing the growing use of these sophisticated algorithms is in the space of Anti-Money Laundering. A survey conducted by Chartis Research found that 48% of the survey responses from financial institutions highlighted the importance of Artificial Intelligence (AI) and Machine Learning (ML) in this regard [1].

The question that follows, is how can artificial intelligence assist organisations with combatting Money Laundering?

Although in essence the ultimate question which  needs to be answered is a binary one – is a particular activity hiding illicit money laundering activity or not – reaching a simple and correct conclusion is in many cases not an easy task. Where do challenges stem from and how can Artificial Intelligence assist?

First and foremost, sophistication exists on both sides of the coin, and it would be naïve and consequently detrimental to think that offenders or fraudsters do not use intelligent techniques to try to conceal their activities. Taking a perspective which is more aligned with an artificial intelligence framework, the typical money laundering techniques of placement, layering and integration may be simplified   as two – first, trying to keep under the radar, avoiding activity that can be classified as anomalous, and second, trying to conceal unnecessary connections, either with other individuals or entities.

With this in mind, we begin to see how artificial intelligence comes into play and how it can help detect, deter or stop altogether fraudsters from reaching their desired goal.

Anomaly detection is one of the most common use cases of artificial intelligence and has already been applied successfully in various industries for finding and identifying outliers to prevent fraud, adversary attacks, and network intrusions that can compromise organisations’ future.

The strength of these techniques when it comes to AML, is that these algorithms  automatically take a holistic view of an individual, including the segment they fall under, their individual background details and activity history, and analyse from all different dimensions whether any activities do not happen  within the ‘normal’ parameters. This composite single view of the individual makes it a more effective approach to identifying anomalies rather than, for example, having single rules which try to analyse different activities separately. Tring to unveil any concealed connections form part of this composite view. Intelligent algorithms can automatically score various internet sources, from professional networks, social networks, news articles, third party databases and official registries to identify interests and links that provide further intelligence to arrive to an accurate evaluation.    

This methodology also aligns more naturally with a risk-based approach. Rather than assessing different events separately, possibly with separate rules, and subsequently taking their individual fragmented outcomes to calculate a single risk score, artificial intelligence algorithms innately consider all the information and collected intelligence holistically to calculate this risk score. Additionally, whether your organisation handles few thousands of transactions daily, or millions, an algorithm can act with the same level of thoroughness on all transactions.

Another aspect which is common in Artificial Intelligence, but which lends itself nicely to this domain, is model calibration. Money laundering is an area where we typically face a moving target – changes in economic activity, regulators’ policies and regulations, political disruptions, criminal activity, major world events, and so on. These events can drive offenders to swiftly change their approaches. When an organisation applies hard rules to monitor transactions, these rules naturally incorporate in themselves the known, but are not as effective when it comes to detecting the unknown patterns. Artificial Intelligence techniques can be designed to detect such changes, automatically re-calibrate, and help organisations stay ahead of the pack, including the offenders.

[1] Chartis. (2017). RiskTech100 RiskTech100 2018

Author

Dr Vince Vella, Chief Technology Officer, Computime Fintech


Share this article
Share on facebook
Facebook
Share on twitter
Twitter
Share on linkedin
LinkedIn
Share on reddit
Reddit
Share on email
Email
Share on print
Print