Summary of the Operate Stage
The Operate stage is the long-running period where an AI capability is live in production, generating consequential outputs and remaining under the canonical governance methodology. It sits between Deploy (go-live) and Sunset (structured retirement) and is where most real-world risk materializes.
Key Activities in Operate
- Telemetry Execution
- Drift monitoring
- Output-quality sampling
- Override-rate and hallucination-rate tracking
- Return on AI (ROAI) measurement
- Risk Register Maintenance
- Entries are updated as materiality, autonomy, or scope change.
- Loss of a Risk Register entry is itself a governance failure.
- AI Council Governance
- Material changes to configuration, tier, audience, or use case require AI Council review.
- Decisions are recorded in the Decision Log (GOV-13).
- Continuous Improvement & Optimization
- Continuous Improvement Cycle (USE-06) and Continuous Optimization Cycle (COT-01) run against live telemetry.
- Findings feed back into configuration, guardrails, training, and documentation.
- Quarterly (or Periodic) Retrospectives
- GOV-15 portfolio sweeps include every Operate-stage capability at its defined frequency.
A capability remains in Operate only while it both (1) produces consequential outputs and (2) is governed under the canonical methodology.