Definition
A model trained on yesterday's patterns can become less accurate when customers, markets, products, fraud tactics, or policies change. That performance shift is model drift.
How it works
Teams compare live inputs and outputs with training baselines, track performance metrics, investigate anomalies, and retrain or adjust the system when drift becomes meaningful.
Why it matters at work
Drift turns previously reliable AI into hidden operational risk. Monitoring is essential for regulated, customer-facing, or high-volume AI systems.
Workplace example
A demand forecasting model becomes less accurate after a supply shock, so planners detect drift and adjust the model before inventory decisions compound.
Frequently Asked Questions
Can model drift be prevented?
Not completely. Drift is a normal production risk. The goal is to detect it early, understand the cause, and respond with retraining, recalibration, or process changes.