Model drift is the gradual decline in a machine learning model’s accuracy over time as real-world data diverges from the data it was trained on, requiring monitoring and retraining.
How it works
Drift takes two main forms: data drift, where the distribution of incoming inputs changes, and concept drift, where the relationship between inputs and the outcome changes. Both cause a model that once performed well to make steadily worse predictions unless it is detected and corrected.
Why it matters for enterprise AI
A model is not a build-once asset — the world it was trained on keeps moving. Detecting drift through continuous monitoring, and retraining before accuracy erodes, is a core reason production AI needs MLOps rather than a one-time deployment.
