What is model drift, and why does it matter in AML monitoring?

Detailed Explanation

Drift is the gap that opens between the world a model was trained on and the world it is scoring today. Customer mix changes, payment rails change, and laundering typologies change deliberately in response to detection. A drifted model degrades quietly. It keeps producing scores that look plausible while its ranking gets steadily worse, so the failure shows up as a slow rise in missed cases rather than an outage. That is why continuous monitoring of the model is a control in its own right: track input distributions, alert rates and outcome rates over time, and set a threshold at which the model is revalidated or retrained rather than waiting for an examination to surface the problem.