By Redko · Updated October 13, 2026
1. Establish the business baseline
- Name the person who owns the workflow and the people using it.
- Record task volume, handling time, typical errors, and the consequences of failure.
- Choose a success measure, a comparison period, and a point at which the project should stop or change.
2. Confirm data and access
- Inventory data sources, content owners, permitted uses, and update frequency.
- Define identity, role and tenant boundaries, retention, deletion, and sensitive data handling.
- Check vendor terms, model data policies, procurement, and any required specialist review.
3. Test the whole task
- Build examples for normal work, ambiguous requests, missing evidence, and unavailable tools.
- Check retrieval, citations, tool inputs, approval steps, and the final user outcome.
- Measure quality, response time, and cost together; do not approve a release from a single demo.
4. Prepare a controlled rollout
- Agree staged access, training, escalation, fallback, and rollback.
- Name the owners of monitoring, incidents, evaluation updates, and model changes.
- Document the implementation, ongoing costs, source access, and maintenance expectations.
Turn the checklist into a scope
Ask each prospective delivery team to identify the evidence they will produce at every stage. A written scope should say what is included, what depends on your team, what counts as acceptance, and how changes affect budget and timing.