How Machine Learning prevents soaring costs in payment screening
Regulatory requirements and the growing trend for instant payments, are forcing bank compliance departments to rethink their systems. They are under increasing pressure when checking payment transactions against embargo lists. The keywords are safety, speed and cost awareness. Unless the process is largely automated, compliance teams are overwhelmed by the number of manual interventions and costs simply explode.
Reduction in false positives despite fuzzy matching
Many regulators also require banks and financial institutions to use fuzzy matching when comparing incoming and outgoing payments against sanctions lists, embargo lists and other blacklists. Fuzzy matching means that risky transactions have to be identified if the string of characters in a name is not only exactly the same but also similar to that found on a sanctions list.
What’s the catch?
Fuzzy matching tends to result in more false positives than exact searches because it inevitably identifies more cases. This is where machine learning comes in as a component of artificial intelligence. A trained machine learning model can optimise the search so that only true positives – so real risks – are reported to the compliance team for manual clarification.
VP Bank Group halves workload, despite fuzzy matching
These were the challenges that confronted the internationally active VP Bank Group. In order to manage the trade-off between cost and risk when using fuzzy matching, the private bank introduced a new payment screening system based on machine learning. The software combines various similarity algorithms to find as many risky transactions as possible (effectiveness) while at the same time finding the least possible transactions that, on closer inspection, turn out to be innocuous (efficiency). The result shows a significant increase in the quality of hits while halving the work involved.
Our current white paper provides further interesting insights into the topic of payment screening with machine learning.