Somewhere in your repairs operation, a piece of software is already making a judgement call. It might be sorting reported jobs by urgency. It might be suggesting a trade, a slot, or a priority code. Most repairs managers could not say precisely how it reached that answer. For a while that did not matter much. It is starting to matter now.

Adoption has moved fast, and governance has not kept pace. A Grant Thornton survey published in June 2026 found that only 13% of construction and real estate leaders believed they could pass an AI governance audit within the next 90 days. Skills gaps hold back around half of AI adoption in affordable housing, and data privacy concerns affect 45% of projects. At the same time, the rules around repairs are tightening. Awaab’s Law Phase 2 starts on 30 November 2026 and puts a statutory clock on a far wider set of HHSRS hazards. When a case reaches the Housing Ombudsman, the question is never “which tool did you use”. It is “why did this job sit for six weeks”. If part of that answer is a prioritisation model nobody in the building can explain, you have a problem that is not really about AI at all.

The fix is not to strip AI out of repairs. It is to insist that anything making a judgement leaves a trail behind it. Good looks like this. The system records what the resident reported, in their own words. It records the evidence that came with it, the photos and the video. It records what priority was set, when, and on what basis. If a vulnerability was flagged, that sits on the record too. A year later, someone who was not there can open the job and follow the reasoning from start to finish. That is not a technology standard. It is the same standard you would apply to a surveyor’s notes.

Alix is built around that record. Residents describe the problem through a structured intake that captures photos and video at the point of report, rather than a free text note typed by whoever picked up the phone. Vulnerabilities are flagged at intake, not discovered three weeks later. Triage puts jobs in priority order using what the resident actually sent, and everything stays in one place alongside the job. When a complaint lands or an Ombudsman case opens, the evidence is already where the job is. Nobody is reconstructing a decision from three systems and somebody’s memory.

The interesting question about AI in repairs is no longer whether it works. It is whether you can show your working. Teams that can will find the next two years considerably easier than teams that cannot. It is worth checking which one you are before November.