Before we use AI to fix public services, can we stop rebuilding the same ones?

I spent yesterday at the techUK Local Public Services Innovation Summit 2026 and one thing kept coming up in different forms: how can councils work beyond their own boundaries and stop inventing everything from scratch?

There was AI, data, procurement, reuse, prevention and shared standards. All things we’ve discussed and argued about in many variations and forms in our team at Telescopic over the years (and years…).

Listening to it, I kept thinking about the organisations on the other side of the council boundary: because local government doesn’t actually deliver all of these services itself – a big part is commissioned out to nonprofits. I leave the whys and political accounting ideas behind this on the side for now.

But from a purely practical implementation point of view – this is where it gets a bit mad.

Imagine a charity delivering victim support under a contract from Essex County Council. Another organisation might provide a similar service commissioned by Cornwall Council. Cardiff Council might commission another organisation to deliver broadly the same kind of support.

Of course they’re different. The local context changes, the people change (although do they really?), referral partners change and each council will have its own requirements (although there you can go down the rabbit hole: why?). But are the underlying processes actually that different?

Someone is referred. Information is collected. Their needs are assessed. People provide support – in person, referrals, notes, calls therapy, other practical help. Things happen over time and somebody needs to record the actions and outcomes.

The council needs to know what happened because it’s paying for the service.

Yet each organisation can end up designing its own processes, choosing and configuring its own systems, creating its own data structures and working out how to report everything back – every council and funding contract requires something slightly different.

I know this because we get brought in to sort some of it out.

We’ll find a CRM configured around the requirements of one contract, spreadsheets that have slowly become systems, slightly different definitions of the same person in four places and staff manually copying information because the systems don’t speak to each other. CEOs that have to reinvent team structures around obscure requires. And usually there’s a perfectly reasonable explanation for how it got there: I mean, yes, the organisation simply had to deliver a service. It just built what it needed at the time.

A nonprofit can spend years working out how a service works in practice. Not how the process diagram said it would work when somebody commissioned it, but what actually happens when real people use it.

They learn which information is more important than others, which referrals go wrong, which questions people don’t understand. They move from freetext fields to defined outcome categories. They know where cases get stuck. Which bits of the reporting are useful and which just exist because, well, they’ve always been there.

Then the contract gets retendered and surprise – another organisation wins it. So suddenly 30 people lose their jobs, maybe some knowledge moves with the service. Some code bits sitting in a CRM nobody else understands. Some obscure configuration in a file sharing cloud service. And quite a lot of it is in people’s heads. The data is all there. And then it isn’t any more.

Because then – we start building again. Usually with a different supplier. Same thing, slightly different variation of people, processes, systems, all done from scratch again.

The GOV.UK Reuse Library is trying to make it easier for government teams to find and reuse patterns and components that already exist. LOTI has been doing something related across London, getting councils to collaborate on technology, data and innovation.

There is an obvious principle underneath these ideas: if somebody has already solved the problem, have a look at what they learnt before you build another version.

I’d like to take that much further into commissioned services. There is a huge amount of knowledge sitting across councils and the nonprofits delivering their services. Some of it will be super specific to an organisation or place. A hell of a lot of it won’t be. So which outcomes tell us something useful? And which fields are we collecting because somebody added them to an Excel reporting template in 2017 and nobody has dared delete them since?

And how do we share and save this info?

That’s where I’d really like us to start reusing: Knowledge, processes, common sense to acknowledge that, actually, we’re not doing something for the first time here…

If it were easy, someone would have done it already I guess. But maybe I’ll start right here with thinking out loud.

What does a homelessness referral need to capture to enable future services to prevent the issue in the first place? How does that data get connected not just with similar non profit organisations, but local governments and other outsourced services? How can we ensure this process is captured in a place where others can find it – and contribute?

How on earth could we collect and use or share data without breaking every single well meant data privacy law? What does a victim support journey tend to look like without giving away very sensitive and potentially dangerous information? What information needs to pass between a council and the organisation delivering the service not just to enable the service but to prevent the need in the first place?

There were some interesting examples of AI in local services today, including work around housing and homelessness and tools to help caseworkers find information, create case notes and manage handovers.

Can we combine information across services and spot something earlier to enable more effective prevention? Can we understand a person’s journey instead of seeing a series of separate interactions with separate organisations? Can we intervene before somebody reaches crisis point?

Yes please, that would be nice.

So let’s start to look at the data underneath it:

Five councils commission similar services from ten different organisations. Each has designed its processes slightly differently and configured its systems around them. Even something as apparently simple as “referral date” can mean different things. Is it when somebody submits a form? When the referral is accepted? When eligibility is confirmed?

Those aren’t minor technical differences if you’re trying to analyse what happened to somebody over time. Multiply that problem across years of information from dozens of services and then try to train an AI model on it.

Good luck.

We can easily see where the problems are in this quest for all seeing data and patterns – this isn’t product recommendation data. We’re talking about homelessness, social care, victim support and other services that can hold a lot of information about people’s lives. If we want to aggregate that information across councils and commissioned providers, how do we keep it safe?

Who needs access? Who controls it? What happens when a contract changes provider? What information actually needs to move across the boundary and what should stay where it is? Who deletes it again?

We then add AI: where is that model hosted? Not just physically but who controls that data in reality? And yes I am talking about mistrusting big private corporations. What happens to the information sent to it? Which jurisdiction is it in? What is it allowed to infer about someone? Who can REALLY access it no matter what their official policies say? There’s a worst case scenario waiting to happen.

And not just access: who is responsible when the model gets the prediction wrong?

We probably can’t solve the fragmentation problem by creating an enormous central pot of sensitive data and hoping the governance catches up later.

Because, let’s face it: centralising the learning doesn’t mean centralising everything else. We’ve seen that the bigger a project is the less likely it ever sees the light of day.

Councils and nonprofits don’t all need the same CRM systems. They don’t need identical processes and they certainly don’t need another enormous monolithic government technology project. But surely we can give organisations a better starting point than a blank piece of paper.

Reusable service patterns. Shared definitions. Starting patterns for data models and standards or Integration patterns. Components that solve common problems. Examples of what has worked elsewhere and, probably more usefully, what hasn’t. How to report coherently. One version all councils can use.

Then adapt those to the service and the people actually using it.

That would also give us a much better foundation for the AI work people like Kat Sexton and others were talking about at this event. Because suddenly there is at least some common meaning underneath the information we’re asking machines to analyse.

The idea of local government looking beyond its organisational boundaries makes sense to me. But those boundaries don’t stop at the next council. They include the nonprofits actually delivering a lot of these services.

If we want data to move across those boundaries, and eventually use it to prevent problems rather than repeatedly dealing with the consequences, we need to sort out what the data means, how it can be shared safely and how much of what we’ve – as an industry in cooperation with our clients – already learnt can be reused.

Before (or rather “while”, given that it’s already happening) we invest in AI across public services, we should invest in the shared foundations that make learning reusable: service patterns, common definitions, safe ways to share relevant data points, and commissioning that preserves rather than discards hard-won knowledge. I almost imagine a public-service franchise model: not one central system, but a tested service model that organisations can adopt, adapt and improve with set boundaries. Shared foundations with local delivery. But that’s a whole other article coming…

So: who’s going to start planning this?