Rule-based filters can detect well-known fraud patterns of payment transactions very effectively. However, for new and dynamic fraud patterns, rules are not that effective. As fraudsters decode the declines and find new ways to perpetrate fraud, the rules become stale. Here we find that AI/ML-based probabilistic fraud filters are more effective as it learns from the labelled data and creates models to detect fraud probabilistically.
RTP is a very nascent channel of payment. Risk Analysts are learning about the patterns as it happens. In the absence of large volumes of labelled data, rules become effective. Nevertheless, as the usage grows and labelled fraud data becomes available, the variation is better learned by AI/ML models. Thus, a combination of rules and AI/ML model becomes effective.
An efficient and effective Fraud Risk Management solution must use a two-pronged approach to detect fraud: (a) define and deploy rules by carefully examining fraud patterns and (b) apply data science and AI/ML techniques to continuously adapt to changing fraud patterns with minimal latency. This dual mode of fraud detection which is now imperative for any reliable FRM solution is a key feature of RS IntelliEdge™ from RS Software.