Leapfrogging Fraud Detection in Banking: Transition from rule-based to AI approach

Traditionally, Fraud detection engines are rule-based and make it difficult to codify new rules immediately. On top of that, it also has a low accuracy level of fraud detection. This results in limited types of fraud that can be detected, affecting the company's reputation and creating reduced operational efficiency. In turn, this negatively impacts customer satisfaction and causes the bank to suffer from fraudulent loss.

To overcome this, KewMann created a comprehensive AI approach to ensure high-accuracy fraud detection by applying proven Machine Learning methods in a near real-time manner. Instead of using the single rule-based approach, the AI approach allows multiple models, which include rule-based, supervised models, unsupervised models, and linked/relationship methodologies. They also used a knowledge graph with network analysis to gain actionable insights for discovering fraudsters faster.

Outcome

- KewMann reduced false alarms by 50% to increase accuracy and achieved near real-time detection with fast data and 30% more types of fraud detected.