The AI Post-Deployment Monitoring Plan
Template
Decide what to watch once an AI system is live, what level of change triggers a review, and who has authority to act.
Signals to watch
- Output quality on a sample reviewed by domain experts.
- Differences in outcomes between user groups where decisions affect people.
- Override, escalation and complaint rates.
- Upstream changes: model versions, system prompts, data sources.
Review triggers
- Set a threshold for each signal that triggers a review.
- Treat a sharp fall in human overrides as a signal to check, not a success.
- Define which events require the impact assessment to be revisited.
Who can act
- Name who is alerted for each trigger.
- List the actions available: pause, roll back, restrict, or accept and record.
- Record who has authority to take each action without further approval.
Review cadence
- Set a fixed review date even when no trigger fires.
- Record each review outcome as evidence for governance and assurance.
Start a conversation
Have a product or AI decision to make?
Useful first calls usually start with one unclear decision, a deadline and a team that needs a practical next move.
Tell us about it