Designed the 0→1 estate-agency product line that puts the agent, the law firm and the seller on the same case — from lead to exchange.
A 15-week build inside a live UK conveyancing platform. Ten workflows, one shared record, five people.
10 systems → 1 · 15 weeks discovery → live · 15+ prototypes, roughly one a week
About the project
Homey ran the legal half of a UK property sale — the conveyancing that begins at sale agreed and ends at completion. Everything that produces that moment happened somewhere else, in estate agencies, on tools Homey had no visibility into. Cases arrived cold, assembled by someone else, as an email.
The Estate Agency CRM was the move to own the other half. Not as a second product that syncs with the first, but as the same case seen from earlier. I led it from an ambiguous market opportunity to a live product line.
The challenge
An estate agent's job ends at sale agreed. A conveyancer's begins there. Between those two moments sits a gap the industry has learned to live with: the legal side spends its first fifteen to twenty days chasing the agent for information the agent already has — the price, the parties, the solicitors, the identity checks, the disclosure, the documents. Not because anyone is careless, but because the two professions have never shared a record. Everything crosses the seam as an attachment in an inbox holding four hundred other cases.
So the design problem wasn't ten features for estate agents. It was this: could two professions with genuinely different jobs work the same record without being forced into each other's workflow? The agent needs speed, triage and marketing tools. The conveyancer needs provenance, accountability and a matter they can open cleanly. One shared surface slows both down. Two synced systems rebuild the seam in software.
What made it hard
Two professions, one roof.
The agency product had to sit inside the mental model conveyancers already used, without forcing agents to think like conveyancers — two different jobs, two different rhythms, one record underneath both.
Regulation expressed as interface, not policy.
UK consumer-protection rules meant the seller's disclosure could never be pre-filled by the agent. Anti-money-laundering rules meant a property could not go to market before every legal owner was identity-verified. Compliance had to become a visible state in the product, not a document in a folder.
Built inside a live platform, not beside it.
Homey already ran real conveyancing cases for real firms. Every new agency workflow had to attach to the existing case record and its existing domain language rather than start a second system.
Fifteen weeks.
Discovery, definition, design, build and release for a full product line, with a five-person pod and a fixed live date.
1 — The structural decision
The obvious build was a CRM that talks to the conveyancing platform through an integration. It would have shipped faster and it would have been wrong — every sync is a place where two copies of the truth can disagree, and disagreement between an agent's file and a solicitor's file is the exact failure the product exists to remove.
So the agency workflows were built onto the existing case record rather than beside it. The property, the parties, the price, the identity checks, the agreed commission and the legal file all hang off one spine. The conveyancer isn't sent the agent's data. They're looking at the same data. Everything downstream — the automatic commission line, the Memorandum of Sale that can't drift, the identity check the law firm can see — is a consequence of that one decision rather than a feature built on top of it.
2 — Mapping the job before the software
Two separate flow sets, deliberately. The first was the ground-level process: what an agent physically does between a valuation call and a live listing, including the parts that happen on a phone in a car park. The second was the software flow. Keeping them apart mattered, because the fastest way to build the wrong operational product is to map the software and assume it describes the work.
The ground-level map is where the ten-system count came from, and where the fifteen-to-twenty-day intake chase stopped being an anecdote and became the thing to design against.
3 — Lifecycles, not forms
Operational software fails at its states, not its screens. A signed agreement is different from a sent one. An approved brochure is different from a generated one. A Memorandum of Sale that has left the building must never change again, however much the live case changes underneath it.
Each workflow was designed as an explicit lifecycle — draft, sent, in progress, completed, voided, failed — and the artefacts carrying legal or commercial weight snapshot at the moment of send. This is the difference between "find the latest attachment" and a record that can tell you what is authoritative and when it became so.
4 — Instruction
Terms of Engagement are usually a Word document, signed in a generic e-sign tool, filed, and then irrelevant. I designed it as the opposite. The agent generates it from the head-office template with case-derived values, counter-signs as the agency, and sends magic links to every legal owner. Nothing is signed until every owner has signed.
Then it does two jobs nothing else in the product could do. It blocks marketing — the property cannot go live until it is signed and every owner is identity-verified. And the commission the seller agreed to is recorded onto the case at that moment, which is the fact that reappears, unedited, on a law firm's completion statement months later.
5 — One declaration, many owners
This is the hardest thing in the product and none of it is visible in a screenshot.
The Seller Enquiry Form is a legal declaration under UK consumer-protection rules, which means the agent may never pre-fill it — only the property-identity section is populated, every disclosure field belongs to the vendor. But a property can have several legal owners, and they are declaring one shared set of facts, not one form each. So a co-owner who edits an answer after someone else has already signed has silently invalidated that signature.
The product handles it explicitly rather than quietly. The edit voids every signature that predates it. Each affected owner is emailed and told which specific questions changed — not "the form was updated" — and asked to review and re-sign. Their dashboard card flips back to in-progress. The agent sees a partially-signed form and can tell exactly who is outstanding and why. Nothing can be changed quietly after signing.
Underneath it, the objective answers — tenure, council tax band, service charge, lease years, EPC — are reconciled against the records Homey already holds, so a mismatch between what a seller declares and what is on file surfaces at instruction rather than four months later in the legal pack.
6 — AI inside the workflow, not beside it
The traditional brochure route runs photos through a phone, copy through ChatGPT, layout through Canva, approval through WhatsApp, and produces a PDF nobody can confidently identify as final. The tools are individually fine. None of them owns the outcome.
The brochure flow infers features and drafts a description from the photos, generates the page layout, and takes conversational refinement — but the agent's description is placed word-for-word and the model is instructed never to paraphrase it. The vendor approves or rejects with a reason from a magic link, and the approved photos, description and highlights become one canonical marketing record on the case. The same content then feeds the portal listing, so the two cannot diverge.
The design principle: AI is allowed to draft, arrange and accelerate. It is not allowed to be the author of record.
7 — From a viewing to a sale agreed
Traditionally a viewing produces feedback in a phone call and a price indication in someone's memory, and then the offer is typed from scratch into a spreadsheet. The buyer gets recreated at every stage.
Here the applicant is a company-scoped record who can view many properties. The viewing carries interest, buying position, feedback, a proposed price and a delta against asking. When an offer arrives, it inherits all of it. Offers made outside the platform — phone, walk-in — enter the same model rather than sitting beside it. On acceptance, the Memorandum of Sale is prefilled from the case, reviewed by the agent, frozen, and delivered to both solicitors in one action.
The agent also gets the property's own argument before they walk in: guide price against asking, area benchmark, last sold price and date, rental yield, nearby schools and transport — so the conversation on the doorstep is informed rather than improvised.
8 — The handover that isn't a handover
This is where the structural decision pays out, and it is the clearest single proof that the two sides share a record rather than exchange copies of one.
At exchange, Homey seeds the draft commission invoice automatically — reading the terms the vendor signed at instruction, the authoritative exchanged price flowing back from the legal side, and the assigned conveyancing fee earner from the law firm's half of the same case. The agent reviews and sends. When the firm opens the completion statement, the commission is already there as a locked line item.
Nobody re-keys it. Nobody can forget it. And the fifteen-to-twenty-day chase at intake has nothing to chase, because there was never a transfer.
Scope of the work
• 10 workflows shipped in v1
• 25+ screens designed
• 15+ coded prototypes
• 6 status lifecycles, each with its own void and failure paths
• 6 magic-link surfaces for sellers and buyers — no app, no account
• 8 third-party integrations designed around
• 9 automated chase routines replacing personal memory
• Per-agency feature toggles, so no workflow could assume the one before it had run
The result
Before Homey, taking a property from valuation call to live listing meant an agent moved between roughly ten systems — a CRM, three research sources, a Word template, an e-sign tool, a KYC portal, ChatGPT, Canva, and a portal back office — and typed the property address five separate times. Every transfer was a chance for a version to fork or a fact to go missing.
The Agency CRM collapses that into one guided chain on one case. The address is captured once at UPRN lookup and never re-entered. The commission the seller agrees to in the Terms of Engagement is the commission that reaches the law firm's completion statement. The description an agent approves in the brochure is the description that publishes to Rightmove.
The conveyancer's first fifteen to twenty days used to go on chasing the agent for information the agent already had. There is nothing to chase now, because there was never a transfer — the law firm opens the same record the agency has been working in since the valuation. The Memorandum of Sale arrives as a structured snapshot. The commission invoice appears on the completion statement by itself.
Rather than claiming an outcome I cannot yet measure, the honest version is the mechanism: the touchpoints that used to depend on someone remembering are now chased by the system, and marketing is blocked rather than trusted.
Shipped end of July 2026. Homey now has a second product line in an adjacent market, with agencies onboarding and the sales team taking it out. I continue to own feature prioritisation, implementation review and product/UAT sign-off.
