Quick Guide · Approval
How to Build an Evidence Pack Before Approving an AI Tool
Collect current supplier evidence, expose the gaps and make approval conditions visible before anyone commits.
The idea to keep
Approve conditions, not a demonstration: gather enough evidence for a defensible decision.
Best for
- Buying or piloting AI tools
- Cross-team approval
- Procurement and governance
What you need
- Primary supplier documents
- Decision owner and criteria
- A record of gaps and assumptions
Assemble the four proof points
1
Name the decision and owner
State what is being approved, for which workflow and by whom.
2
Collect current primary evidence
Request product, privacy, security, support, export and pricing evidence.
3
Record gaps and assumptions
Mark claims that are unverified, outdated, conditional or future-dependent.
4
Set approval and exit triggers
Define what must be resolved before approval and what triggers review later.
Example
Four-card evidence pack
Product claim
Capability statement linked to the tested workflow.
Privacy and data
Processing terms, retention and data boundaries.
Support and SLA
Support route, response expectation and incident process.
Export and exit
Usable export, deletion route and exit trigger.
Check before you use it
- Evidence is current and attributable.
- Decision owner is named.
- Gaps are visible.
- Exit trigger recorded.
Common mistake
Collecting supplier material without linking it to the approval question.
Nova9 view
Evidence makes approval accountable
A clear pack lets another person understand why the tool was approved and when to revisit it.
See the NCSC secure AI guidelines.
