Payment and transaction screening helps detect sanctions violations across multiple regulatory regimes. Banks have setup large screening programmes, consisting of policies, processes, systems and people, to detect and manage sanctions violations. These large screening units were needed essentially to manually examine the alerts that are usually false positives.
In screening parlance, a false positive occurs when a screening system flags the contents of a transaction i.e., the payment message, to match with a sanctioned entity, when in reality, it is not a match. Similarly, a false negative happens when the screening system misses to detect a true sanctions violation. Look-back analysis of past transactions often reveals the missed violations, which can require investigations by internal screening unit as well as external personnel.
The results from large screening units also helped banks fix data quality and sourcing issues. Some banks used guidance frameworks such as BCBS 239 to implement data governance. Master data management and fuzzy matching improved the effectiveness of screening filters resulting in the capture of incremental sanctions violations. White lists and hit reducing rules helped identify false positives by weeding out spurious matches.
Despite the advances in accuracy of detection, banks’ screening units are still facing challenges in terms of high volumes of false positives and missed detection of sanctions violations and the associated fraud and other financial crimes. The figure below depicts a typical state of the art payment transactions screening flow. The lower portion of the diagram contains what is possible in terms of improving the speed (efficiency) and accuracy (effectiveness) of financial crimes detection. Authors of this article as well as other forward-looking banks and financial institutions are now successfully experimenting with these new Artificial Intelligence (AI), Machine Learning, and cognitive computing technologies.
Robotic process automation (RPA) appears to be the most commonly implemented emerging technology to replace the repetitive human tasks. In the FCC world, RPA is being used to dispose false positives and complex cases involving fraud or other financial crimes (where the nature of breach can be determined with a high degree of certainty, based on historical data). RPA is an AI technique based on human experience. It mimics human actions for a much speedy and error-free execution.
Authors also implemented Machine Learning, Deep Learning (another AI method), and statistical models such as Random Forests, graph databases, and auto-encoders to detect and codify new fraud and violation patterns, thereby improving the overall effectiveness (number of true positives/ number of violations). These new models have also been able to improve the speed of detection by finding new indicators for old fraud patterns.
We are in a new era of financial crimes detection where in Machine Learning, AI, and cognitive computing methods are improving financial crimes detection and compliance. Banks and financial institution that invest in these improved methods will not only be progressing to a higher level in detecting financial crimes but also improve operational efficiency and forge stronger relationships with the right customers.