What Is Model Drift?

Model drift happens when an AI or machine learning model becomes less reliable because the real world it is working with has changed since it was trained.

A model might originally be very good at spotting patterns in the information it receives, but those patterns don’t necessarily stay the same forever. Customer behaviour changes, new types of fraud appear, markets move and the way people use products and services evolves.

If the model doesn’t keep up with those changes, its predictions or decisions can gradually become less accurate.

What Does Model Drift Look Like?

Imagine an AI system trained to detect suspicious card transactions based on patterns from previous years. If the way people shop or the techniques used by fraudsters change significantly, some of the patterns the AI originally learned may no longer be as useful.

The model hasn’t necessarily broken. The world around it has moved on.

That’s why organisations need to keep checking how AI models perform after they’re introduced. A model that worked well when it was launched won’t necessarily continue working in the same way as the world around it changes.