Do you have a plan for AI model retirements?

AI models don't stay available forever. On Azure, a model version is retired about 18 months after its launch. On that day the deployment either switches to a different model by itself or stops working. Either way, the app changes without a release.

That's why every project should know when each of its models retires and move to the replacement on its own schedule.

Don't find out from an outage

❌ Figure: Bad example - Nobody knows which version is deployed or when it retires

✅ Figure: Good example - Every deployment has a date, an upgrade setting and a backlog item

Steps to manage AI model retirements

1. List your deployments - Record each deployment's model, version and retirement date
2. Put the date in the backlog - Add a PBI three months before each retirement, so there is time to test
3. Get the notice as a team - Microsoft's email only goes to subscription owners. Add a Service Health alert for Azure OpenAI Service and send it to the project's Teams channel
4. Choose the upgrade setting on purpose - For production, use "Once the current version expires": the version only changes at retirement, and the app keeps running if the plan slips. Provisioned deployments are never upgraded automatically
5. Test the replacement first - Run your evals against the new model before you switch