How a Bank Cut Check Fraud by 90 Percent with Machine Learning

The Operations Podcast with Fexingo: Process, People, and Profit in Modern Business · Jul 8, 2026 · 6m · transcript

Episode 101 of The Operations Podcast dives into one of the most expensive operational headaches in banking: check fraud. Lucas and Luna examine how a regional bank used machine learning to flag fraudulent checks in real time, slashing losses from $12 million to under $1.2 million in 18 months. They break down the specific model design — a gradient boosting classifier trained on 400 features per check — and the workflow changes that made it stick: routing suspect checks to a three-person review team instead of mass rejection. Along the way, they discuss the trade-off between false positives…