Anti-money laundering (AML) groups have historically lagged their fraud counterparts in the use of advanced analytics, due in large part to the heavy burden of model risk management processes borne by AML departments. However, mounting international sanctions, increasing payment volume, and painful enforcement actions are dictating that financial services firms look to new and better ways to ensure AML compliance. Robotic process automation (RPA) and machine learning technologies are now being embraced by AML teams around the globe as they look for ways to improve both detection and operational efficiency.
RPA is the use of advanced analytic technologies to handle high-volume, repeatable tasks that traditionally require human intervention. AML is rife with examples of these types of tasks. I recently sat on a panel with the head of AML compliance for a large North American wealth management firm. He said that the AML analysts on his team currently spend 60% of their time on data gathering and 20% on actual analysis. His goal is to flip that ratio using RPA over the next two years. Many other large financial institutions are following suit, looking to RPA to create efficiency while also making regulators happy.
Machine learning encompasses analytic techniques that can identify patterns of behavior through iterative optimization. It has been widely used in fraud detection for years and is now finally starting to be used for AML, as financial firms are able to demonstrate to regulators that these models can provide explainable outcomes while improving detection and reducing false positives. I have spoken with a number of large global financial services firms that have deployed advanced analytics for a variety of AML use cases, and many more that have it on the near-term roadmap.
AML represents an environment for machine learning techniques, which excel in recognizing anomalous patterns amid large datasets. Recognizing the value that these analytics can provide, many regulators have started hiring their own data scientists for their audit and review teams to be sure they are asking the right questions and not impeding what could be substantial progress in providing more meaningful AML outcomes.
The growing volume and complexity of payments and banking is such that legacy approaches that rely solely on rules or models that are only updated once every year or two just don’t work anymore, nor does throwing bodies at the resultant mountain of false positives. AML is entering an exciting new phase in which machine learning and RPA can substantially improve the status quo (for everyone except the bad guys, that is).