AI in UK Social Housing

Three separate authorities, with no coordination between them, reach the same conclusion. The Regulator of Social Housing says breaches are caused by weak data and an incomplete understanding of the condition of tenants' homes. The Housing Ombudsman says data integrity is pivotal to the outcome of a complaint. And the sector's most-quoted AI success story records its own accuracy falling from 97% to around 70% when the underlying data is incomplete. The technology is not the constraint. The record of what a landlord knows, and when it knew it, is the constraint — and from 30 November 2026 that becomes a legal question rather than an operational one.

There is a second, less comfortable story. The Housing Ombudsman has already stated publicly that it expects more use of AI by both landlords and residents in the complaints process. Complaints are arriving pre-screened by no-win-no-fee advisers, citing specific legislation and the Ombudsman's own code. The asymmetry has flipped: in this sector, the people making claims are better equipped than the people receiving them.

A child whose bedroom window was boarded up for four years, the family worried about excess cold. A household living two years with bin bags covering a hole in the ceiling, with a risk of asbestos. A resident without functioning heating and hot water for three years. Every one of these, published by the Housing Ombudsman as real cases, falls under a hazard category that becomes legally timebound in under four months from Phase 2 — and most landlords cannot currently evidence, for a given property, when they first became aware.

Sixteen pages, structured to be read by a stakeholder who is new to AI but knows the sector well:

• A note on the numbers — what is included, what was excluded, and why
• The wider market and the specific pressure point: Awaab's Law and the "becoming aware" clock
• The conduct picture from the Housing Ombudsman, and the regulator's own findings on data quality
• The threat side — AI used against landlords by residents and their advisers, and why that is not the same story as fraud
• The current UK legal and regulatory position, including the Data (Use and Access) Act 2025
• Where AI actually is in the sector today, and the flagship case study handled honestly
• What can be done today, and how the opportunity develops over the next three years
• Benefit realisation — what to measure, and where most benefit cases quietly break
• An eight-workstream way in

Every figure carries a source. The "98% accuracy" claim attached to the sector's best-known AI case study is not used as published — the vendor's own case study says 70–97%, conditional on data quality, and that condition is the actual finding. Widely-circulated statistics that could not be traced to a named regulator, ombudsman, or landlord's own statement are left out, and the exclusion is stated in the document itself.

A senior stakeholder at a housing association or local authority landlord who has just picked up an AI programme covering repairs, damp, or building safety — or is about to be asked to sign one off — and wants a plain, sourced account of where things actually stand before the vendors get to the conversation.