There's a debate happening across local government right now that will sound very familiar if you work in social housing. It goes like this: should we fix our foundations before we adopt AI, or can we do both at the same time?

At the Socitm Presidents Conference last month — the annual gathering of technology leaders across UK public services — a room full of senior digital professionals voted on exactly this question. The majority chose "fix the foundations first" or "do both simultaneously." Almost nobody chose "accelerate AI regardless."

That instinct is right. But it can also become a reason to do nothing.

One speaker put it clearly: every council has a digital strategy claiming AI will deliver financial savings. But AI has never delivered savings. Leadership decisions do. The same is true in social housing. The technology doesn't fix anything on its own. The question is whether you're pointing it at the right problem.

The organisations that are getting results aren't the ones with the biggest vision. They're the ones that started small and specific. Merton Council had a FOI team of three people managing 2,000 requests a year, an 78% compliance rate, and a formal warning from the ICO. Their AI sprint didn't start with a strategy — it started with pain. A proof of concept built in one afternoon reduced processing time from seven hours per request to fifteen minutes. Accuracy is now approaching 95%. The IG team wrote the prompts themselves.

Liverpool City Council's principle — "buy first, find out second is the failure mode" — is the lesson behind every AI implementation that doesn't land. You start with the pain point, not the product. Then you pick the smallest thing you can actually measure.

For social housing repairs teams, the pain points aren't hard to identify. Backlogs that don't shrink despite everyone working flat out. Operatives arriving at jobs without enough information to fix it first time. Leaders who can't see the full picture because it's spread across five systems and a dozen spreadsheets. Resident complaints that could have been prevented if someone had spotted the urgency sooner.

This is where practical AI earns its place. Not a platform that promises to "transform" everything, but a tool that solves one specific problem well enough that people actually use it. Alix is built for exactly this: giving repairs leaders visibility across their whole job list, flagging what's genuinely urgent, enabling remote diagnosis before committing an operative, and keeping everything — photos, vulnerability information, decision trails — in one place. Not AI for its own sake. AI for the backlog.

The Socitm conference also highlighted something worth sitting with: only 14% of AI proof-of-concepts ever reach scale. The ones that do have something in common — they were chosen because they hurt, they were built with the people doing the work, and they had measurable stakes from day one.

Start where it hurts. Build with your team. Measure it. That's the model.