MLOps & Infrastructure · intermediate · concept 215 of 311 · Updated Nov 4, 2026
Model & Data Drift
Drift is what happens when the world changes after a model is trained, so its performance slowly falls without any error being raised. Data drift means the inputs changed; concept drift means the relationship between inputs and the right answer changed. Monitoring for it and deciding when to retrain is a large part of keeping a model useful.
L8 · Production and scaleDevelopersTech leadersLeadershipProduct leadersQA and testingSEO and content
Key terms
Data driftConcept driftMonitoringRetrainingPopulation stability
Learn these first
Where you meet it in the real world
Fraud models after new attack patterns, demand forecasts after market shifts, vendors silently updating LLMs
Guides and articles
What is data drift? ↗
Evidently AI
What is concept drift? ↗
Evidently AI