RICS Moves to Frame AI Valuations as Regulators Race to Catch AVM Adoption

The timing is awkward in the way that most property regulation is awkward: the market moved, and the standards body is running to catch it. RICS put its first-edition global practice guidance on AI in real estate valuation out to public consultation in Q2 2026, with final publication pencilled in for later this year. The document is designed to give RICS-registered valuers and regulated firms a framework for using AI responsibly inside professional workflows, not as a bolt-on curiosity but as a formal part of how valuations get done.

That framework is overdue. Automated valuation models are not experimental in the UK mortgage market; they are the market. Hometrack's AVM alone processes around 50 million valuations a year and is used by 17 of the top 20 UK mortgage lenders, covering close to 75% of the mortgage market by volume. The technology can produce an instant price estimate for a typical UK home in under three seconds, and roughly 80% of those estimates land within 10% of a chartered surveyor's figure. The accuracy question, though, is where the professional tension lives.

What the Manchester Data Actually Shows

Researchers at the University of Manchester published an AI system claiming over 96% accuracy in house price prediction, against the 70–85% typically associated with traditional comparable-sales methods. That delta matters less than it sounds. Traditional valuations are built for low-transaction postcodes, listed buildings, and properties where the physical fabric tells a story that postcode-level data cannot. AVMs are frankly terrible at all three. An AI cannot inspect a property, identify a damp problem, judge build quality, or price a Grade II listed barn conversion in a village with four comparable sales in the last decade.

RICS is threading this needle. The guidance is framed not as a prohibition or a blessing but as a governance structure: professional judgement, transparency, and accountability must sit above the model output. The model is a starting anchor, not a final figure. That is already how the better-run lenders use AVMs. The risk is further down the intermediary chain, where an estate agent or online auction platform serves up an AVM result to a seller as if it were a surveyor's opinion.

The DIY Landlord Problem

The valuation debate sits inside a broader industry-wide data gap that proptech has been circling for years. A July 2026 survey of more than 340 self-managing landlords, conducted by proptech platform August, found that 92% still ran their lettings on spreadsheets or paper rather than on dedicated software. That is the same cohort making pricing decisions on hundreds of thousands of rental properties without AVM tools, AI or otherwise.

For the AI valuation story, this matters because the residential rental market is structurally opaque. Sold-price data is registered with HM Land Registry; agreed rents are not, at any comparable granularity. The Renters' Rights Act has accelerated institutional interest in transparency here, and the Fyma/GAA Living partnership announced earlier in 2026 explicitly targets the BTR data gap by pairing computer vision analytics with BTR advisory intelligence to generate denser occupancy and demand signals.

Trademark Signals: Builders Pulling Back

Brand-registration activity in NICE Class 37 (construction and repair services) offers a useful leading indicator of sector confidence. According to AIBD analysis of UK Intellectual Property Office trademark data, Class 37 filings reached just 599 in Q3 2026, a fall of 65.8% against the prior period. Trademark registrations typically trail commercial intent by six to twelve months, so a collapse of this magnitude in the construction and maintenance services class suggests forward pipeline confidence across development and refurbishment work is thinning sharply. That has direct implications for the valuation market: fewer new-build completions and refurbishment schemes means fewer comparables for AVMs to anchor against, which degrades model accuracy precisely when lenders are leaning on it most.

The Governance Gap

JLL's 2025 Global Real Estate Technology Survey found that 88% of commercial real estate investors and owners had already started AI pilots. Yet only 5% reported achieving most of their AI programme goals, and more than 60% admitted they remained strategically, organisationally, and technically unprepared for scaled implementation beyond pilots. The residential sector is at an earlier stage of institutional adoption, but the direction is the same.

The RICS guidance, when published, will not close the commercial-to-residential gap on its own. What it establishes is something the AVM market has lacked: a named professional standard that chartered surveyors can point to when a client, a lender, or a regulator asks how the AI was used and what override process existed. In a market where Hometrack's model processes the equivalent of the entire UK housing stock roughly once a year, that accountability layer is not a bureaucratic nicety. It is structural risk management.

The consultation window closes this year. The industry has until then to argue about where the human stops and the model starts.