What happens once a model goes live?
Deploying a model is the start of its working life, not the end of the work.
What is it?
Putting a model into service is the start of the work, not the end. The world changes, the data change, and a model that tested well can degrade over time. So it has to be monitored, updated and, if necessary, withdrawn.
An everyday example
A card-fraud model was trained on last year’s purchases. When fraudsters change tactics, it starts missing new patterns until it is updated.
Why does it matter?
Most practical problems with AI appear after deployment, not before. Without monitoring, the users are the first to notice the mistakes.
A common misconception
That a model which once passed its tests is reliable for ever.
In more detail
What is model monitoring?
Regularly measuring quality, errors, delays and risks after the model goes live.
What does “data drift” mean?
Real-world data can drift away from the training data over time, so a model that was good last year is not necessarily as good today.
What is MLOps?
The practices and tools for developing, deploying, measuring and maintaining models.
Can a model be withdrawn?
Yes. If it causes harm or does not meet the standards, it has to be fixed, restricted or switched off.
Remember: A model needs watching for its whole life cycle.
Reliable sources
Last reviewed 1 November 2026. An explainer, not professional, legal or medical advice. For rules and figures that change, check the official version.