Fraud Detection Isn't a Machine Learning Problem

This article argues that fraud detection should be viewed as a decision-support system rather than a standalone machine learning problem. While model metrics such as precision, recall, and AUC remain important, the author emphasizes that real-world fraud prevention depends on operational workflows, behavioral analytics, threshold management, explainability, monitoring, and analyst usability. The central thesis is that the most effective fraud systems optimize for decision quality and business ou