Skip to content
How we assess the evidence

How we assess AI case evidence

The library helps readers discover operating patterns. It does not treat a customer story as proof that the same result will occur elsewhere.

Sources checked

1

Company source

Published by the organisation using the system.

2

Vendor-published case

Published by a technology provider about a named customer.

3

Independent source

Published by a party that did not sell or implement the system.

4

Methodfield interpretation

Our explanation of how the documented workflow may transfer to another context.

5

Limitation

Information needed for evaluation that the source does not provide.

Minimum publication standard

A published case names the organisation, describes the business problem and workflow, links to a source, distinguishes reported results from interpretation, states material unknowns, and identifies the human-control boundary.

Outcome wording rule

We use ‘the source reports’ unless a result has independent support. We do not convert a qualitative statement into a number or attribute a multi-step outcome to one isolated AI task.

Problem before product

Compare the reviewed cases

Cases are examples, not prescriptions. A useful pattern still needs to fit your data, workflow, controls, budget, and operating environment.

Back to all cases