TL;DR

Worcestershire County Council has published eight months of operating figures for an agent that vets evidence submitted with concessionary travel pass applications. It handled 2,498 applications, checked 7,494 uploaded documents, and turned back 643 photographs, 364 unsuitable proofs of address and 286 dates of birth that did not stand up. A human still makes the decision; the agent works upstream of them.

A narrow job, done properly

The problem the council set out to fix was mundane and expensive: staff time spent eyeballing documents, deciding whether they met criteria, and ringing residents to explain what was wrong. Each of those calls is a round trip that delays an application.

The agent intercepts the failure at submission. It can tell that a photograph does not meet the specification, that the wrong document has been offered as proof of address, or that a date of birth is invalid — at the moment the resident is still on the page, rather than days later. Jo Hilditch, who manages the digital delivery team, is careful about what that does and does not represent: “The goal was never to replace good service design or human judgment with AI. It was to make work faster, safer, and smarter.”

A second tool, a validator for school attendance fines, is credited with saving as much as 200 hours a year on document checking.

Governance shaped the technology

The constraint worth noting for other UK public bodies is data residency. Because the council must process data in the UK only, some newer models are unavailable to it, GPT-5 among them. Rather than treat that as a blocker, it embedded AI inside defined applications with known users, controlled data sources and a clear business purpose, which lets it pick a different model per use case and watch token spend against limits.

That containment is the strategy: governed applications rather than assorted tools spreading through the organisation unobserved.

Looking forward

Set against the national picture — Scotland’s public sector AI guidance and Merseyside Police’s Bobbi chatbot — what distinguishes Worcestershire is that it published counts rather than ambitions. Hilditch’s advice to other councils inverts the usual order: start from the service problem, get governance and data in place, and pick a handful of use cases that can show something.