Ed Molyneux argued last week that two AI shifts have compressed the Open Property Coalition’s roadmap. This week he sets out the four-layer stack a conveyancing firm actually needs to take advantage of those shifts, and why most firms are buying the wrong layer first.
If last week’s column persuaded you that the agentic transition in conveyancing is 12-18 months away, the obvious next question is what to do about it. And the obvious wrong answer is to buy more AI.
Most firms shopping for AI tools right now are buying the bit that goes on top – a chatbot, a drafting assistant, a summariser. These can be genuinely useful. But on their own they are the icing on a cake nobody has yet baked. The firms that compound an advantage in the next 18 months are the ones building four layers in the right order.
Layer one: data that machines can read
Almost no AI conversation in property starts where it should — with the data. Most firms still receive answers as free-text in property information forms, attached as PDFs and forwarded by email. Anything that arrives as a scanned image, a phone-call note, or a ‘see below’ reply is friction that an agent cannot dispatch on its own.
The fix is structured data flowing on a known schema – in property, the Property Data Trust Framework (PDTF). Where a piece of information arrives in PDTF-compliant form, with provenance and a lifecycle event attached, an agent can act on it. Where it does not, a human has to translate. Every PDTF-shaped touchpoint you add removes the need for translation later.
Layer two: trusted skills, not raw models
The second layer is the specialist skills the model uses to do regulated work.
A general-purpose model asked to calculate stamp duty will sometimes get it right and sometimes get it confidently wrong. The same model wired to a deterministic SDLT skill always gets it right. The point is not that the model is too dumb. The point is that an SDLT calculation is not the kind of question a language model should be answering on its own at all.
The same applies to lender part one pre-screening, lease eligibility, AML triggers, building regulations checks and several dozen other things in the conveyancing day-to-day. These are skills, not prompts. They should be curated by people who know the work and trusted by the profession. We have published the Moverly Conveyancing Toolkit as open source for exactly this reason, but the point holds whether you use ours or build your own.
Layer three: hooks into your existing stack
Layers one and two are useless if the AI lives in a separate browser tab. The third layer is where the agent picks up its instructions and where it writes its output – your case management system, your inbox, your WhatsApp, your document store.
Look at your CMS, whether that is Leap, Proclaim, Osprey, Tikit or anything else. Ask whether an agent could read open matters from it and write draft enquiries back. Look at the inbox you use for client communications. Ask whether an agent could classify incoming messages and propose responses for human review. Look at where documents arrive. Ask whether they get classified, extracted and tagged automatically, or whether someone is still file-naming them.
You do not need to integrate everything. You need one good hook this quarter, and another next quarter.
Layer four: a human oversight model
The fourth layer is the one most firms skip entirely. It is the explicit set of rules about what an agent does autonomously, what it surfaces for review, and what it absolutely never does without a human.
A worked example: your agent drafts initial enquiry letters from a flagged risk report; the fee earner reviews and sends. Your agent classifies incoming emails and routes them to the right matter; the fee earner can override on a morning briefing. Your agent prepares a first-pass certificate of title; a partner reviews before signing. Never the other way round.
Without rules like these written down, agents drift and people lose trust. With them, your firm has a clear language for what AI is and is not doing, which is also exactly what your regulator and your PII insurer want to see.
What to do this quarter
Pick one layer that is missing or weak in your firm, and put a name against it. Not all four. One. Most firms already have something in layer two and almost nothing in the other three. The biggest leverage point right now is layer three – one CMS or email integration that turns AI from a parallel sandbox into something that genuinely works inside the firm.
The roadmap is the destination. This stack is how you skate there.
About the author
Ed Molyneux is co-founder and CTO of Moverly, the property intelligence platform working with LMS and Connells Group to bring structured, verified data to property transactions. He is the architect of the Property Data Trust Framework (PDTF), the open standard for machine-readable property data now being adopted across the industry. Ed writes about AI, property data infrastructure, and the future of conveyancing.
