Across five years at the ICB — four in the quality team before moving into the System Coordination Centre, then a year there — I kept running into the same gap. The NHS generates enormous amounts of operational data, but data literacy across the system is uneven. Analysts often don’t have clinical experience to judge which numbers actually matter; clinicians and administrators often don’t have the grounding in data interpretation to know what to ask analysts for, or how to read what comes back. I sat unusually across both — clinical background, comfortable with data — and people noticed. I was regularly the person asked to take a piece of narrative (“providers are telling us X”) and check whether the data actually backed it up.
Copilot’s enterprise access — organisational data alongside the open web — made a particular kind of work faster: pulling local narrative and data together, then using deep research to bring in the national picture, related work elsewhere, and evidence that could support or challenge what the local narrative was claiming. Three examples below use that same underlying method for genuinely different problems.
Mental health bed capacity
One recurring issue on the system calls was mental health patients waiting — sometimes for days — in emergency departments for a bed, or on wards waiting for a mental health placement. The obvious operational question is whether there’s actually a shortage of beds, or whether something else is driving the wait. I decided to try to answer it properly rather than accept the narrative at face value.
I started locally: whatever reporting and documentation on mental health bed capacity and ED waits I could access. Access itself was part of the problem — information across the organisation was siloed by team, and I could only see as much as I could because I sat across both the System Coordination Centre and the quality team’s information. Even with that reach, it wouldn’t have been everything the organisation held.
I then used Copilot’s deep research capability to look at the national picture: how mental health capacity and waits were being reported nationally, and whether other local areas were seeing the same pattern. Alongside that I went looking for national datasets myself — bed availability data, in particular — while also asking the AI to find the same.
The data didn’t triangulate cleanly. Bed availability was reported nationally by provider, not by Integrated Care Board — the geography didn’t match. Waiting times were logged on a different geographic basis again. To bring the two together meant making explicit assumptions about how provider-level bed availability mapped onto the ICB-level waiting-time picture, rather than treating the numbers as directly comparable. Untangling that — different reporting units, different time periods, working out what could and couldn’t be fairly compared — was the kind of detail that would have taken a lot longer to work through by hand.
Once the local documentation, the national data, and the operational narrative from the system calls were reconciled against each other, it became a report. I shared it with colleagues, and my manager passed it on to commissioners.
Patient transport
Non-emergency patient transport came up regularly as an issue on the system calls — a complex, largely human-operated logistics system, and one people kept flagging as a source of delay and friction. It looked, on the face of it, like a strong candidate for AI-assisted efficiency gains, though the genuinely hard part of that is the data governance and confidentiality involved in patient transport logistics, which is a large part of why progress on it had been slow.
I generated research into this — the kind of thing AI is being used for elsewhere in transport logistics, and what that might mean for how patient transport could be run differently. I shared that with my manager, and we passed it on to the colleagues responsible for the patient transport contract. It seemed clear to me that AI ought to be something explicitly considered in that contract — something the service should be thinking about over the next few years, not left implicit.
Turning national policy into a local briefing
A different application of the same underlying skill: national NHS England policy and guidance documents — I did this a couple of times, particularly around urgent care and acute medical admissions guidance released towards the end of my time at the ICB — tend to be dry and abstract, written for a general audience rather than for any one system.
I took new policy documents as they came out and used Copilot to generate an executive briefing: what was actually new in the document, what had changed relative to the guidance it replaced, and — the part that mattered most — what it meant for us locally. Being able to connect a national document to our actual local picture, and specifically to flag the areas we hadn’t already started work on, was the useful part. A summary of the document alone wouldn’t have told anyone that.
Sources: written from my own recollection. There’s no chat record for this work, because it predates my exported chat history and involved NHS operational data, so the underlying documents, data and the exact prompts aren’t reproduced here.