The hype is true! Artificial Intelligence (AI) has the potential to be an incredibly powerful tool. But, as with any great tool, to maximize its impact you have to know how and when to use it.
In money laundering and terrorist financing (ML/TF) detection, we are all very familiar with traditional approaches that use AI. For example, in transaction monitoring, financial institutions commonly use AI-based systems to alert investigators to suspicious transactions -- which is a vast improvement over trying to perform manual reviews of all transactions. However, these systems have a false positive rate estimated between 90% and 95%: that is, of every 100 transactions alerted, less than 5-10 of those turn out to be, after significant investigation and analysis, truly suspicious.
It is the performance of these systems have caused many experienced professionals to question the true usefulness of AI. Not only are false positive rates high, but investigators still have manual and repetitive work to perform: after the AI fires, they need to comb through myriad data sources, including disparate internal databases and external data sources, to determine if the alerted transaction is truly indicative of money laundering or terrorist financing.
A New View of AI to Combat Money Laundering
The reality is that AI performing only transaction monitoring will have significant numbers of false positives. The training data sets are highly imbalanced, as true suspicious activity is extremely rare relative to non-suspicious activity, and the data is only partially labeled, since there are many ML/TF accounts that remain undetected. However, when considering the ML/TF problem from a holistic viewpoint, it turns out that there are a number of ways in which applying AI differently can both better use current transaction monitoring results as well as improve the results themselves for the future.
Innovative approaches using AI focus on the areas in which AI can add the most value:
Performing the more mundane day-to-day tasks, like aggregating the massive amounts of available data
Finding & automating new patterns in the investigator workflow
Improvements in these areas free the investigators and analysts to focus on more complex future threats and make far better risk assessments. And, as the identification of suspicious activity becomes better, the data sets used for training the AI become better, bootstrapping a more effective and efficient system for suspicious activity detection.
No More Black Boxes
Modern and innovative AI is best implemented as a highly adaptive system, but with the appropriate controls and monitoring to ensure that it is performing as expected. Too often, systems are packaged as black boxes: producing KYC scores with little explanation of what that score means, or red flagging for a transaction because it originated in, for example, a known suspicious geography -- well, suspicious 10 years ago.
These systems can be difficult to understand, and impossible or expensive to update and customize. This negatively impacts the capability to address new and emerging threats, or even just to test new ideas from analysts and investigators. A great AI system allows its users to quickly update the system, to run tests and experiments, and provides clear, human-readable reasons behind the decisions it makes.
Artificial Intelligence can be a massive game changer – but more and more financial institutions are realizing that to harness its true power, it must be implemented intelligently and focused on where it can make the greatest impact. Financial institutions are finding that those who invest into this effort will be the global leaders of the future.
At Clovis Technologies, we're developing innovative technologies to unlock more of the potential of AI. Contact us to learn more about smart strategies to implement AI technology at your financial institution.