Planning window
Ongoing; propose a monthly operating review
Planning estimate, not a delivery promiseWho to involve
Service and process owners · Finance · Security/privacy · Users · Executive sponsor
Monitor business and quality measures
Track accepted outputs, corrections, time including review, user adoption and cost per completed task. Investigate movement in the baseline instead of attributing every improvement to AI.
Maintain governance
Keep an AI system register, approved-tool policy, owners, processor records and incident routes. Re-evaluate data access, legal classification and supplier changes.
Scale by evidence
Test model or prompt changes against the evaluation set. Expand to a new process or department only after checking its own data, economics and responsibilities. Retire systems that no longer justify their cost.
What you leave with
- KPI and cost dashboard
- AI register and policy review
- Prioritised improvement and retirement backlog
Common mistakes
- Watching token costs without task outcomes
- Expanding permissions with no review
- Updating models without regression evaluation