Most housing associations have had the same meeting by now. Someone's built a long list of things AI could help with: triage, void turnaround, complaints, content, contractor scheduling. The list keeps growing. Nobody picks anything from it, because someone always asks what happens with the edge cases, and the meeting moves on to the next agenda item instead.

That pattern came up clearly at a recent Pantheon webinar on AI adoption, hosted by Lucy with guests David Bishop and Alex Langley, though their audience was local councils rather than housing providers. Most councils, they said, are still drafting AI policy and have got no further than Copilot. The blocker isn't usually a lack of options. It's too many of them, combined with real discomfort about anything beyond "human in the loop." Autonomous decisions feel risky in a sector already under regulatory scrutiny, so caution wins by default, and caution rarely ships anything.

What the panel found worked was narrower than the wishlist suggests. Email triage running roughly ten times cheaper than a human doing the same sift, flagging safeguarding issues and prioritising the backlog automatically. A voice AI agent fielding calls on a single phone line, deflecting around 30% of them, for about £15k across a three-month pilot. Nothing sector-wide. One use case, one measurable result, then a decision about whether to go further.

That's close to how Alix's own customers get started, and for the same reason. Teams that try to automate triage across every repair type at once usually stall, for the same reasons the councils on that panel stalled: too many unknowns to govern all at once. The ones that succeed pick one entry point, say damp and mould reports coming in through WhatsApp, and let Alix structure and prioritise those first: flagging urgency, capturing evidence, surfacing the cases that need a human decision fastest. Once that's trusted, the second use case is a much easier conversation, because it's now a known quantity rather than a leap of faith.

If your AI list has more items on it than pilots running, that's not a strategy problem. It's usually just the list winning the meeting again. Pick the one closest to a phone ringing or a form landing in an inbox, and start there.