Use a completion checklist
GOV.UK's service guidance describes acceptance criteria as checks that a user need has been met, with supporting evidence. A pilot brief can use the same approach to make acceptance explicit.
GOV.UK: acceptance criteriaA defined result your team can check.
An early-access delivery model in development for a specific piece of work. Agree the output, completion checks and reviewer before work begins, then assess the result against that brief.
Early-access services. Scope, integrations and pricing agreed before work starts.
Example pilot: reconcile an agreed batch of records. Set the checks before work starts so your team can make a clear acceptance decision.
Specify the input set, required output and checks that determine acceptance. A pilot could produce a source-linked document review, a reconciliation report or a software change with agreed verification.
A named reviewer accepts the work. Missing evidence, unresolved exceptions or failed checks are recorded and handled through the agreed correction process.
A completed report is a deliverable. Reducing the effort needed to prepare and review it is a separate business result. Record the current process before comparing the pilot.
Include review time, corrections and operating costs in that comparison. Pricing may be fixed for the pilot or tied to accepted units; the proposal defines the unit, exclusions and rework terms.
These references inform pilot planning. The delivery tools and responsibilities are agreed for your task.
GOV.UK's service guidance describes acceptance criteria as checks that a user need has been met, with supporting evidence. A pilot brief can use the same approach to make acceptance explicit.
GOV.UK: acceptance criteriaGOV.UK recommends combining performance measures with user research and other data. Task completion and time taken can be compared over repeated assessments; output count alone does not establish value.
GOV.UK: measuring service successDocumentation checked 15 October 2026.
Each pilot needs a clear agreement on access, data handling, approval and exceptions.
Related guidance: OWASP AI agent security and NIST generative AI risk management. Controls and responsibilities are agreed for each engagement.
The service model starts with a deliverable your team can accept. Savings, revenue and other business effects need separate measures and evidence. They are not guaranteed by completing a deliverable.
Record the failed checks and apply the correction or escalation process in the pilot agreement. Review windows, rework and excluded cases are agreed before delivery.
Your team names the accountable reviewer. Automated checks and evidence support that person's decision; an agent's own assessment does not establish acceptance.
Bring one deliverable, an example of acceptable work and the person responsible for signing it off. We will assess the access, integration and review requirements before proposing a pilot.