Model Drift

Model drift is the gradual or sudden degradation of a model's performance in production caused by changes in input data, target distribution, or operating context.

What is Model Drift?

Drift comes in several flavours: covariate shift (inputs change), concept drift (the relationship between input and label changes), and prior shift (label frequencies change). Detection methods include monitoring input distributions, output distributions, calibration, and business KPIs. Response options are retraining, re-thresholding, or rolling back to a previous model. Drift monitoring is required for high-risk AI under the EU AI Act post-market monitoring obligations.

How does Model Drift apply to enterprise AI?

Enterprise ML systems trained on 2024 data may drift in 2026 as customer behaviour, product mix, or regulatory definitions shift. Without drift monitoring, the team learns about the problem from customers or regulators.