What should a pilot request include so the first sample can be judged quickly?
A checklist covering process, condition types, ranked exceptions, deliverable shape, formats, intended use, success criteria and a business reviewer.
Put these items in a pilot request so the first sample can be judged quickly, and keep the request short enough to fit on one page.
- One named process and domain, with a plain description of the decision and the work that follows it. - The condition types and states you want represented, and any states your own systems use that differ from the usual ones. - Three or four exception patterns ranked by how much they cost you when an agent gets them wrong. - The deliverable shape: dataset only, task pack with ground truth, or environment with tools and reset. - Target formats, identifier style and time zone handling for your pipeline, so nothing needs remapping on arrival. - Your intended use of the output, so licence terms can be discussed against it from the start. - Written success criteria, including the baselines you will run and what separation between them you expect. - A named reviewer from the business side who can say whether the policies and ground truth look sensible.
A request that names process, exceptions, formats and success criteria lets both sides judge the first sample against something concrete.
- Audience:
- Agent developers
- Content type:
- Checklist
- Buyer question:
- Scope and getting started
Content release: r-a220eb6e099b243b
Related questions
- What does a first pilot review look like for a referral-processing agent team?
- How is generated conditional-approval material licensed, and what limits apply?
- How should I scope a first pilot for conditional-approval material?
- How do I wire a pilot sample into my agent harness for a first run?
- What should I ask for in a first sample or pilot?
- What should I ask about licensing and permitted use before starting?