HomeNeutral EvaluationUse casesTechnology, data & AI
TN/USE/13
Edition 2027 · effective 1 January 2027
Technology, data & AI · Is this AI system fit for its use
The model card says it works. Does it work for your users?
Before deploying AI in hiring, lending, health or public services, an independent evaluation of claims, performance across groups, and human oversight.
§ 01The evaluation question
What the evaluator is asked.
“Within an agreed evidence and testing scope, does this AI system perform as claimed for its intended use, and what remains unassured?”
(The question)
Commissioned by. Board, risk committee, regulator-facing teams
When. Before deployment and yearly after
Usual protocol. B · Single-commissioner decision review
§ 02Specialist mandate required
A general neutral credential is not enough here.
Why
Regulated or specialist ground
An AI-system assessment is assurance-heavy technical work. It needs defined access to evidence and the system, an agreed testing scope and qualified technical experts. It gives limited assurance against the stated use, and is not certification or regulatory conformity assessment.
What the scope review checks
- The specialist qualification or licence the question requires
- Whether a statutory or regulated process must be used instead
- Who must be heard, and their procedural rights
- What the evaluation will expressly not cover
§ 03What is examined
The evidence, and the output.
The evidence
Material reviewed, and what was not available, is always stated.
- Claims made by the vendor or team
- Performance on representative data
- Differences across groups
- Human oversight and appeal routes
A fitness-for-use view with conditions
A reasoned view. The decision stays with you.
- It will not: certifying legal compliance with a specific AI law
§ 04Depth and time
Usually: Evaluation Panel.
Two or three evaluators from different disciplines. Typical time: Three to six weeks.
§ 05Your version
Every case is different. Frame yours.
§ 06Related