What to maintain every month after an AI go-live

Short answer: After go-live, review system health and work context every month, not only uptime. Separate model/data, service, security, access and recovery, then record the decision.

A monthly health report

1. Service: availability, latency, queue and actual incidents.
2. Data: input volume, missing fields, drift signals and document/image quality.
3. Model: version, evaluation sample, false positives/negatives and new limits.
4. Security: dependencies/CVEs, access review, secret rotation and support sessions.
5. Recovery: latest backup, restore test, rollback package and ready owner.

Numbers must come from a named system. Do not make a dashboard that makes tested look like supported. Read the Maintenance Agreement and Trust boundaries.

When to retrain or stop

Retrain when new data has approved rights and an agreed evaluation method, not because one number changed. Stop or roll back when acceptance is out and there is no evidence that the fix is safe.

Limitations

A report without an owner, review date, baseline or sample cannot support a decision. Work outside scope must not be called included by default.