Where the AI runs
On-premise, hybrid or public cloud — recorded per entry, because the deployment model decides whether a use case is even admissible in your environment.
A register you can check, not a directory you have to trust.
FindAUseCase.com records real AI implementations for B2B buyers — with the fields a European buyer actually needs and a publish gate that keeps thin evidence out.
What every entry records
Four fields, kept on every published entry. They are the reason a shortlist can start here instead of with a research project.
Which regulations apply
The frameworks that touch each implementation, tagged per entry.
The source behind the claim
Numbers are written exactly as the source states them — never rounded up, never annualised, never re-based. Open the source and check the number yourself.
A verification that holds
Every checkable claim tested against the fetched source, and one unsupported figure fails the whole entry. Nothing publishes on a template.
Source-verifiedPublish gate: a passing verification, still matching the entry's content
How verification works
The publish gate is deliberately boring: named criteria, applied the same way to every entry, and checked by the server rather than argued with.
Entries are drafted from public sources — vendor case studies, conference talks, press coverage, engineering blogs. Drafting is machine-assisted, and so is the checking: a separate pass tries to refute every claim against the source it cites, recording each number as supported, unsupported or not found.
Publishing is then decided by the server, not by whatever wrote the entry. It requires a verification that passed, whose content hash still matches the entry — edit the text after it was checked and it must be checked again — plus a minimum evidence band. I set that bar and audit the register; I do not approve entries one by one, because a rubber stamp would be worse than none.
That gate is new, and it is worth being exact about what it does not cover: the entries published before August 2026 predate it. Those were checked against their sources by hand — a pass that found thirteen entries whose numbers did not match their source, and unpublished them. Each entry's page says which of the two it went through, and the refresh stage is working back through the older ones.
High confidence requires a named company, a described technology stack and at least one quantified outcome from the source. An entry that cannot meet the bar stays unpublished, and a published entry whose claims stop holding up comes down. Unverifiable entries are unpublished, not softened.
Found something wrong? Every page has a feedback control, and corrections outrank additions in the review queue.
Start from the register.
Free to browse. Filter by sector, capability and deployment model, and open the source behind any claim.