What the Home Buying and Selling Reform Roadmap Builds for AI

The Home Buying and Selling Reform Roadmap, published by MHCLG on 19 June, gives AI a modest role in the future of Home Buying and Selling in the UK. Government recognises its potential to automate routine conveyancing tasks such as document classification, triage and data extraction, while preserving human judgement for decisions, and will work with the Digital Property Market Steering Group next year to establish standards for its appropriate use. Most lenders will read those lines, file them under conveyancing, and move on.

That would be a misread. The measures that matter most for AI in lending never mention it.

AI in a mortgage business is only as good as the data it runs on. Today that data arrives late in the transaction, and much of it arrives unverified. The roadmap changes both conditions. Read it as data policy and the AI consequences come into focus.

Four measures, read as data policy

Four measures carry the weight for lenders: upfront sales packs, streamlined and reusable AML, binding conditional contracts, and the digitalisation of land data. Taken in sequence, they change when data arrives, who has to check it, how borrowers behave, and what software can read. Each deserves a look through an AI lens.

Sales packs move data upstream of the offer. Take timing first. Legislation planned for this Parliament will require a standardised sales pack, including searches and a property condition report, before a property is listed. Voluntary versions start this year. For AI, this is the significant change, because most decisioning models start work when the application arrives, and that is when the data does. A verified pack at listing means the property side of a lending decision exists before there is a buyer. The offer stops being the starting gun for data. For a lender, the practical gain lands at application: property facts already verified, searches already run. The work shifts from assembling evidence to assessing it.

The caveat is coverage. Voluntary first means partial for years, so the near-term design question is provenance. Models need to know which cases carry verified pack data and treat the rest differently. The roadmap is explicit that forthcoming guidance will define what upfront information needs in order to support lender acceptance. Lenders who engage with that guidance work will shape the data their own models consume.

Reusable AML changes the question compliance AI answers. The second measure is about trust rather than timing. The roadmap backs identity and AML information gathered once, to a high standard, at the earliest possible stage, then relied on by others, supported by Digital Verification Services certified against the DVS trust framework. For an AI system, this flips the task. The question stops being “is this person who they claim to be” and becomes “can we rely on a check someone else performed”. Reliance is a problem of provenance and trust signals rather than document reading, and it is a regulatory question as much as a technical one, because reliance has to be defensible to a supervisor. That supervisor is changing: the FCA is taking over as the single professional services supervisor for AML, replacing the Solicitors Regulation Authority. The live test is already running. Coadjute, backed by Lloyds Banking Group, NatWest and Nationwide, has been accepted into the FCA’s AI Live Testing programme to test an AI-native AML platform for property transactions, and the programme will inform a forthcoming good and poor practice report for AI in financial services. The direction is compliance AI that reads trust metadata, watched by the regulator while it works in production.

Binding contracts will quietly age your fall-through models. The third measure reaches further than it first appears, because it changes the behaviour the data describes. Around one in three transactions fall through, on MHCLG’s figures. Any lender running fall-through prediction has trained it on years of behaviour under rules where walking away costs nothing. The roadmap commits to legislation requiring binding conditional contracts, with penalty fees for withdrawal without good reason, brought into force once sales packs are embedded, bringing the rest of the UK closer to Scotland. When the cost of walking away changes, the behaviour that predicts walking away changes with it.

Here is the uncomfortable implication: every fall-through model in the market today is trained on behaviour this legislation is designed to end. The models will not fail loudly. Scores will drift and rankings will quietly degrade, while the training data describes a market that no longer exists. None of this happens this year; binding contracts sit at the end of the sequence, deliberately. The decision that lands now is architectural. Teams that already monitor drift and retrain against a moving baseline will treat the transition as routine. The exposure sits with models built once, validated once, and left to run.

A machine-readable register makes title work an automation candidate. The last measure is the quietest, and it may prove the most useful. HM Land Registry will restructure the register into a machine-interpretable format by 2030, and government will work with HMLR to explore the reuse of machine-readable information with lenders to simplify remortgaging. A register that software can read directly turns title checking from a document problem into a data problem. The remortgage book is the obvious test bed: high volume, repetitive, and named in the roadmap as the use case. HMLR is also scaling what it calls the responsible use of AI inside its own operations, which says something about where the registry expects this to go.

Two questions for the next planning cycle

So where does this leave the AI the roadmap does talk about? The government’s framing keeps humans on decisions and points AI at routine work. In the consultation, 2% of respondents advised against increasing AI’s role in conveyancing, pointing to the need to retain a human element in one of life’s biggest events. The response is standards rather than prohibition: DPMSG standards next year, and the AI Growth Lab, whose applications open for legal services and conveyancing firms later this summer. That matches the sequencing case we made earlier this year. Narrow, reliable AI first. Decisions stay human until the foundations earn something more.

Two questions are worth taking into the next planning cycle. Can your models consume verified data created outside your organisation? And which of your models are trained on behaviour the roadmap is about to change? Both are answerable now, on your own estate, before a single measure takes effect.

Sources: MHCLG, Home Buying and Selling Reform Roadmap, 19 June 2026 (gov.uk). FCA AI Live Testing second cohort, April 2026, with coverage in Today’s Conveyancer and The Intermediary. HM Land Registry Strategy 2025+.