AI operations · 9 minute read
Operate production AI as a business service
A service model for quality, incidents, change, cost, suppliers and value after release.
- Operations
- Operate
- Global
- Operate
Decision lens
Who owns AI behaviour, dependencies, incidents and business value after the project team leaves?
Models, prompts, knowledge, policies, tools and providers change. Production AI needs service ownership that can see quality, control, cost and value together.
Define the service boundary
Set the accountable owner, supported workflows, dependencies, service levels, support windows and supplier responsibilities.
- Name business and technical owners
- Map failure dependencies
- Agree service and value baselines
Observe behaviour and operation together
Monitor workflow quality, human interventions, exceptions, drift, access, latency, cost and availability in one operating view.
- Use business-aligned thresholds
- Track knowledge freshness
- Connect alerts to controlled response
Control every material change
Evaluate and approve changes to models, prompts, knowledge, tools and policies, with release evidence and verified rollback.
- Classify material changes
- Re-test affected controls
- Maintain recovery evidence
Practical checklist
Evidence to bring into the decision.
- 01Service definition
- 02Operational telemetry
- 03Incident and rollback plan
- 04Change control
- 05Cost and value review
Continue with evidence
Connect the perspective to an operating method.
Operating scope and service responsibility are engagement-specific. Illustrative telemetry does not represent a live client service.