Reinforcement Learning for Next-Gen AML: From Rules to Dynamic Decisioning

Author(s): Shenggang Li

Originally published on Towards AI.

A Practical Guide Combining Causal RL, and Thompson-Sampling for Watch-List Prioritization and Peer-Group Outlier Detection

Anti-Money Laundering (AML) operations are facing a dual challenge: growing regulatory pressure and increasing transaction complexity. Traditional rule-based systems — such as fixed watch-list filters and static thresholds — are becoming costly and ineffective, generating high false-positive rates and missing sophisticated patterns like smurfing chains or dormant-to-active account spikes. Meanwhile, banks are under pressure to balance compliance with operational efficiency, ensuring every alert and every Suspicious Activity Report (SAR) brings measurable value.

This paper introduces a reinforcement learning (RL)–driven framework for AML, using Causal RL and Thompson-Sampling bandits, enhanced by unsupervised clustering techniques like HDBSCAN. It shifts the paradigm from reactive rules to a proactive learning model, dynamically improving decision-making to decrease false positive alerts and optimize investigation efforts, thereby balancing risk with operational efficiency while maintaining compliance with regulatory standards.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI