# AI Workflow Reliability Brief

Use this brief before an AI-enabled workflow becomes a team dependency. Keep the answers concrete enough for an
operator to run the workflow, an owner to inspect it, and a delivery partner to define what ongoing support means.

## Workflow contract

- Business workflow and decision being supported:
- Named operating owner:
- Executive sponsor:
- In-scope users, systems, and data:
- Human decision or approval that remains in the loop:
- Fallback path when the workflow fails or confidence is low:

## Quality bar

- Evaluation set or representative test cases:
- Minimum acceptable quality or success rate:
- Known failure modes and prohibited outputs:
- Reviewer and escalation path:
- Acceptance evidence and measurement period:

## Production boundary

- Status: pilot / limited production / production / retired
- Release date and current version:
- Rollback or disable path:
- Audit log or system-of-record destination:
- Access, retention, and data-classification controls:

## Reliability service

- Evaluation cadence:
- Drift checks and trigger for re-evaluation:
- Incident severity levels and response expectations:
- Included change capacity:
- Out-of-scope requests and change-order path:
- Sunset, handoff, or renewal decision date:

## Cost and adoption

- Model, inference, and third-party usage cost boundary:
- Usage cap, tier, or BYOK arrangement:
- Adoption measure and denominator:
- Time, quality, or cycle-time measure:
- Higher-leverage work expected from time returned:

## Handoff checklist

- [ ] Owner can run the workflow without the builder present
- [ ] Evaluation set and quality bar are stored with the workflow
- [ ] Monitoring, incident, and escalation paths are tested
- [ ] Documentation and rollback instructions are current
- [ ] Usage costs and limits are visible
- [ ] Next review or sunset decision is on the calendar