Model Drift

Model drift occurs when an AI system’s performance changes after deployment because the world no longer matches its training and evaluation conditions. Customer behavior, sensors, language, policies, markets, or upstream data pipelines may shift over time. Data drift describes changes in input distributions, while concept drift means the relationship between inputs and desired outcomes has changed. Monitoring can track feature distributions, confidence, error rates, and business outcomes, but labels may arrive slowly or not at all. Responses include recalibration, retraining, new rules, rollback, or human review. Teams need thresholds and ownership in advance so a detected change leads to action rather than another unattended dashboard alert.