What is the go/no-go decision?
The decision is a documented approval, conditional approval, pause or rejection for a defined AI use case. It should identify what was assessed, the evidence relied on, the residual risks accepted, the person with authority to accept them and the conditions that must remain true.
UK Government procurement guidance recommends impact assessment at the start of the process and review at key decision points or after substantial design changes. It also calls for ongoing evaluation, knowledge transfer and lifecycle management rather than treating award as the final control.
Separate commercial award from production deployment
A contract can be awarded before every implementation detail is complete. Production use should still wait until configuration, integration, acceptance tests, user controls and outstanding treatments meet the agreed conditions.
Record two decisions where necessary: an award decision based on the proposed solution and contractual commitments, followed by a deployment decision based on the configured service and observed test evidence. This prevents commercial momentum from silently becoming operational approval.
The nine-part evidence pack
- Decision scope. The use case, users, affected people, intended benefit, boundaries and prohibited uses.
- Impact assessment. Human, legal, operational, security, equality and socio-economic consequences.
- Supplier claim register. Material claims, source evidence, accountable reviewer and verification status.
- Data and model evidence. Provenance, rights, quality, representativeness, model dependencies and known limitations.
- Performance evidence. Metrics, test conditions, population coverage, failure modes and acceptable thresholds.
- Control evidence. Human oversight, access, security, logging, transparency, incident and redress arrangements.
- Residual-risk decision. Risks remaining after controls, owner, rationale, expiry and review trigger.
- Contract and exit controls. Change notices, audit evidence, service reporting, remedies, portability and termination support.
- Deployment conditions. Acceptance tests, training, operating owner, monitoring plan and unresolved actions.
How should supplier claims be tested?
Prioritise claims that affect scoring, safety, fairness, cost, service levels or regulatory posture. Request the original test report or data behind performance statements and record whether the evidence reflects the buyer's intended users, language, geography and operating conditions.
Where a claim cannot be independently verified before award, mark it as unverified and decide whether a contractual commitment, acceptance test or deployment condition provides enough protection. Do not convert marketing language into an approved fact by copying it into the decision paper.
What should the decision record say?
Use one of four outcomes: go, when evidence and controls meet the threshold; conditional go, when named actions must be completed before deployment; pause, when material evidence is missing; or no-go, when the residual risk or solution fit is unacceptable.
Record the decision date, decision-maker, evidence index, dissenting views, conditions, residual risks, monitoring owner and next review. Link each condition to a due date and objective closure evidence.
Frequently asked questions
What is an AI procurement go/no-go decision?
It is a documented approval, conditional approval, pause or rejection based on the intended use, evidence reviewed, residual risks, contract protections and readiness to operate the AI system.
What evidence should be reviewed before buying AI?
Buyers should review the use case, impact assessment, supplier claims, data and model evidence, performance and limitation tests, security and oversight controls, residual risks, contract terms and deployment plan.
Should procurement approval also authorise deployment?
Not automatically. Commercial award and production deployment can be separate gates, particularly when configuration, integration, acceptance testing or risk treatments remain incomplete.
A practical next step
Start with the AI supplier evidence guide and the free Procurement Readiness Check. For a material purchase, talk to AI Act Ready about structuring the evidence review and approval record.
Last reviewed 11 November 2026. This article provides general information, not legal advice.