From problem to prototype: what the Bield Council did with Freedom of Information

Every Bield engagement starts with a problem worth solving, so it seemed fair to put our own method through the same test. The problem we chose was Freedom of Information. Local authorities and other public sector organisations handle a steady and growing stream of requests. Scottish public bodies received more than 98,000 information requests in 2024-25, each with a 20 working day legal deadline, and in most organisations one to three people cover information governance. Every request takes officer time to research, draft and check.

Our starting assumption was that AI was the problem. People are now using AI tools to write their requests, and the obvious idea was a system that could spot them. We worked it through with the FOI team at a local authority, and put the idea in front of the Bield Council. It did what it is built to do: it challenged us.

The question it kept coming back to was simple. Where does officer time actually go? The honest answer from the FOI team was that AI-written requests were a nuisance, not the bottleneck. The real cost was repetition: officers researching and drafting answers to questions the organisation had already answered and published.

So we changed the problem. ClearFOI's core became matching each new request against what the organisation has already disclosed, so an officer can see quickly whether the answer already exists. AI helps with the matching and the first draft. People make the decision and own the response. Spotting likely AI-written requests stayed in, but as a supporting signal rather than the headline.

Then data first. We built the prototype on safe test data, then brought in real, published disclosure data. The workflow came out considerably shorter. In our model, officer time on a typical AI-generated request falls from around 4.2 hours to around 1.5. Reading, classifying and routing drops from 45 minutes to none, and response drafting from 75 minutes to 20. The final review stays exactly as it was, because a person still approves every response. These are modelled estimates rather than a measured result. It is still a prototype, and we say so. What it has proved is that the method works.

The twist is that the AI question we parked has come back. In May the Information Commissioner's Office published guidance for public bodies on AI-generated FOI requests, describing requests that are long, repetitive and full of misdirected demands. In October, an elected member in Edinburgh asked for a report on the pressure AI-generated complaints are putting on local authority staff. Because we built for the real problem first, the AI signal is already there for when public sector organisations are ready to use it.

The lesson applies to any business. The problem you walk in with is rarely the one worth solving, and building for the wrong one is the most expensive mistake in AI adoption. Test the problem before you build anything. That is what the Bield Council is for, and it is why every Bield engagement starts with problems, not tools.

ClearFOI is a separate product, at clearfoi.com.

Source references: Scottish Information Commissioner, 2024-25 Annual Report (more than 98,000 information requests). Information Commissioner's Office, guidance for public authorities on AI-generated FOI requests, May 2026, reported by Local Government Lawyer. FutureScot, on AI-generated complaints in Edinburgh, October 2026.