Model drift is the gradual decline in an AI system's accuracy because conditions have changed from those it learned from. The model has not broken; the world has moved on.
It happens when customer behaviour, products, prices, weather patterns or data formats change. A demand forecast trained before a new competitor opened, a menu change or a pandemic-style shock may keep predicting the old pattern. Drift can also occur when a data source changes quietly, such as a new product code format.
Retailers manage drift by monitoring results against reality, such as forecast versus actual sales, comparing performance monthly, and having a clear process to retrain, adjust or switch off a tool. Ask vendors how they monitor and update their models and what alerts you will receive.
The pitfall is set-and-forget. A tool that worked brilliantly at launch can slowly drift into being wrong while everyone trusts it. Seasonal businesses should compare like with like, and major changes should trigger a review. Keeping a simple record of when prices, menus or opening hours changed makes it much easier to tell drift from ordinary variation.
It matters because stale AI quietly costs money through overordering, understocking or poor recommendations. Regular checks protect the return on your investment.