# Ding Dong — Complete Build Specification & Research Pack **A working end-to-end specification for a UK residential property transaction platform.** Version 1.0 · August 2026 · Released freely for the Ding Dong founders to use as they wish. --- ## How to use this document This is written to be handed to an AI coding assistant (Claude, or similar) as the single source of truth for building the application. Give it this whole file and say: > "Read this specification end to end. Build the application it describes, starting with the > data model and the transaction state machine, then the buyer app, then the staff back-office. > Simulate every external API at the boundary described so the product is fully demonstrable > before any real integration exists." Everything here has been validated against a working prototype that was built and demonstrated: consumer app, marketing site and staff back-office, with a complete purchase journey from property discovery to Land Registry registration. **Two principles that made the prototype work, and should govern the build:** 1. **Simulate integrations at the exact boundary where the real one will connect.** Every external system (agent CRMs, lenders, Land Registry, searches, identity checks, stamp duty) is behind an adapter with a defined interface. The demo returns realistic fake data; the production build swaps the adapter. Nothing else in the app changes. 2. **The transaction is an event stream, not a status field.** Every fact about a purchase is an immutable event. The buyer's timeline, the staff case file, and the audit trail are all just different views of the same events. This is also what makes the product defensible to a regulator. --- # PART 1 — THE PRODUCT ## 1.1 What it is A platform that runs a UK home purchase end to end: discovery, viewings, offers and negotiation, mortgage, conveyancing, completion, move-in, and ongoing ownership. The buyer does the enjoyable 5% — swiping, viewing, deciding — and the platform does the other 95%. The commercial insight, borne out by every UK precedent: **you do not make money from discovery, you make money from the transaction.** Discovery is customer acquisition cost. Mortgage, legal, survey, insurance and move-in services are the revenue. Build accordingly. ## 1.2 The ten-stage journey Dream → Discover → Ding & Dong → View → Fair Negotiation → Buy → Move → Own → Grow → Sell, and back to Dream. The platform's advantage is that it never hands the customer off; the same record follows them from first swipe to resale years later. ## 1.3 Target outcomes | Metric | UK market today | Platform target | |---|---|---| | Offer accepted → completion | 10–16+ weeks | 5–8 weeks (chain permitting) | | Fall-through rate | ~25–29% | under 10% | | Buyer admin time | dozens of hours, 5+ providers | under 2 hours, one app | | Remortgage | weeks | about 1 week | | Chasing done by the buyer | constant | none | Fall-through economics justify the entire build: with over a million transactions a year and roughly £1,000+ of sunk cost per failed purchase, reliability is the product. --- # PART 2 — TECHNICAL ARCHITECTURE ## 2.1 Stack used in the prototype - **Backend:** Laravel (PHP), MySQL or SQLite, Redis, queue workers - **Frontend:** Livewire + Tailwind CSS, as an installable Progressive Web App - **Staff back-office:** same application, separate layout and route group - Any equivalent stack works. What matters is the domain model below, not the framework. ## 2.2 Structure A modular monolith with clear domains: Discovery, Listings, Parties, Mortgage, Legal, Marketplace, Ownership, Billing, Intelligence. Split into services only when scale or regulatory isolation demands it — the likely first split is Legal, for ring-fencing. ## 2.3 The integration hub Every external system sits behind an adapter implementing a domain port: `ListingFeedPort`, `SourcingPort`, `LenderPort`, `SearchOrderPort`, `LandRegistryPort`, `IdentityPort`, `EsignPort`, `OpenBankingPort`, `UtilityPort`, `SdltPort` Rules: inbound webhooks verified and idempotent; outbound calls queued with backoff and circuit breakers; every failure lands in a staff "exception inbox" rather than failing silently; nightly reconciliation for every integration because webhooks alone are never trustworthy; full redacted request/response archive per case for disputes. ## 2.4 Data model **users** — id, name, email, role (buyer/staff), household_name, profile (JSON: buying stage, household, work locations, lifestyle tags, budget min/max, minimum bedrooms, commute anchor), finances (JSON: verified flag, joint income, deposit, gifted funds, buying power, credit status, open banking connection, monthly commitments) **estate_agents** — name, branch, initials, brand colour, rating, review count, feed_status, feed_type (RTDF / BLM-SFTP / CRM API), crm (which software they use), average days to sell, percentage of asking price achieved, phone, email **listings** — estate_agent_id, uprn, address_line, area, town, postcode, lat, lng, price, beds, baths, receptions, type, tenure, epc_rating, floor_area_sqm, council_tax_band, description, features (JSON), images (JSON), material_info (JSON: completeness flag, missing fields, parts A/B/C), area_scores (JSON), sold_comparables (JSON), status (live / quarantined / sstc / sold), quarantine_reason **interactions** — user_id, listing_id, action (ding / dong / superding / viewed) *Critical: only ding/dong/superding remove a property from the swipe deck. Merely viewing it must not, or users lose homes they were still considering.* **viewings** — user_id, listing_id, slot_at, status, booked_via, feedback (JSON) **offers** — user_id, listing_id, amount, counter_amount, agreed_amount, status (draft / submitted / relayed / countered / accepted / declined), evidence (JSON) **negotiation_messages** — offer_id, from (buyer / mediator / seller_side / system), body **transactions** — user_id, listing_id, offer_id, state, agreed_price, target_completion, parties (JSON), chain (JSON), fall_through_risk **transaction_events** — transaction_id, type (milestone / api / chase / document / note), title, body, actor, icon, meta (JSON), created_at. **This table is the heart of the system.** **mortgage_products** — lender, lender_colour, name, initial_rate, aprc, fix_type, fix_months, fee, max_ltv, api_enabled, features (JSON) **mortgage_applications** — user_id, transaction_id, mortgage_product_id, status (dip / submitted / valuation / offer_issued), dip_amount, loan_amount, deposit, term_years, monthly_payment, submitted_via **documents** — transaction_id, user_id, kind, title, sign_method (none / esign / qes), status (draft / ready / awaiting_signature / signed), signature_name, signed_at, meta (JSON) **partners** — category (solicitor / surveyor / removals / utilities / broadband / insurance / photographer / cleaning), name, rating, review count, price_from, blurb, api_badge **move_services** — transaction_id, category, status (suggested / arranged / skipped), partner_id **aml_checks** — user_id, type (identity / source_of_funds / pep_sanctions), status (clear / review / failed / pending), provider, meta (JSON) **app_notifications** — user_id, title, body, icon, link, read ## 2.5 The transaction state machine Eleven states, each with a completion percentage driving the buyer's progress bar: | State | Label | % | |---|---|---| | offer_accepted | Offer accepted | 8 | | instructed | Solicitor instructed | 16 | | searches | Searches ordered | 28 | | survey | Survey & valuation | 40 | | mortgage_offered | Mortgage offer issued | 52 | | enquiries | Enquiries | 62 | | report | Report on title | 72 | | signing | Contracts signing | 80 | | exchanged | Exchanged | 90 | | completed | Completed — keys day | 97 | | registered | Registered at Land Registry | 100 | Entering a state emits its events, generates its documents, and pushes a notification. In production these are triggered by webhooks from lenders, search providers, the Land Registry and the conveyancer's case management system. In the demo, a staff "advance stage" button fires exactly the same code path — which is why the demo is honest rather than theatre. **Events emitted per state** (title — body — actor): - **offer_accepted:** "Offer accepted — £X" (memorandum of sale issued to all parties); "ID & AML verified for all parties" (verified to Land Registry standard, shared with both solicitors so nobody asks twice). Generates: Memorandum of Sale. - **instructed:** "Solicitor instructed" (client care pack e-signed in app); "Title documents pulled" (official copies of the register and title plan retrieved in seconds). Generates: Client Care Letter, Property Information Pack. - **searches:** "Searches ordered — day one" (local land charges returned instantly where the council is digital; local authority, drainage and water, environmental and coal mining ordered as one bundle); "Coal mining search included" (standard in former coalfield areas — ordered automatically so it never causes a late surprise). - **survey:** "Survey booked & completed" (RICS Level 2, report digitised, amber items summarised); "Search results back" (all returned and summarised). Generates: Search Results, Survey Summary. - **mortgage_offered:** "Mortgage offer issued" (checked automatically against the contract pack); "Offer → conveyancer handoff" (lender instructs the legal team via panel rails, Certificate of Title scheduled). Generates: Mortgage Offer. - **enquiries:** "Enquiries raised & answered" (drafted from the pack by AI, approved by the solicitor, most answered same-day from the seller's digital property pack); "Chasing: seller solicitor" (chased automatically, escalates to a human after 48 hours). - **report:** "Report on title — plain English". Generates: Report on Title. - **signing:** "Contracts ready to sign" (contract e-signed; transfer deed signed by Qualified Electronic Signature — no witness, no printer). Generates: Contract of Sale, TR1 Transfer Deed. - **exchanged:** "EXCHANGED" (deposit transferred, completion date fixed); "Priority search + buildings insurance" (Land Registry priority secured, insurance live from exchange). Also creates the move-in service records. - **completed:** "COMPLETED — collect your keys" (funds sent against the Certificate of Title, receipt confirmed, keys released); "SDLT filed same day". Generates: SDLT Return, Completion Statement. - **registered:** "Registered at HM Land Registry" (title updated, deeds in the vault forever). --- # PART 3 — THE BUYER APP Installable PWA, mobile-first, five bottom tabs: Discover, Saved, Journey, Messages, Finance. ## 3.1 Onboarding — "Dream" Two-minute conversational intake, never a long form: buying stage, household, work locations and commute tolerance, lifestyle preferences as tappable chips, budget band, deposit status. Output: an initial preference profile plus an honest affordability band using a soft credit check that leaves no footprint. Only show homes they could actually buy. ## 3.2 Discover — the swipe deck The card shows: hero photo with **left/right arrows and dot indicators to browse all photos**, price, match score badge, beds/type/area strip, address, agent, and the top reasons for the score. **Gesture behaviour (get this right — it is the first thing anyone tries):** - Drag the card horizontally; it follows the finger with a slight rotation - A green "DING" stamp fades in when dragging right, an orange "DONG" stamp when dragging left, opacity proportional to distance - Past a ~90px threshold on release, the card flies off and the decision is recorded - Below the threshold it springs back with no action - A tap (no drag) opens the property — track whether movement occurred and suppress the navigation if it did - Photo arrows must sit **outside** any navigation link, or the router will intercept the tap - Buttons remain as an accessible alternative: Dong (✕), Super-Ding (⭐, tells the agent the buyer is serious), Ding (💚) - Give each card a unique key so animation state does not leak onto the next card ## 3.3 Match scoring Transparent, weighted, and — critically — **self-explaining**. Every score shows its reasons. Start with explicit features, not a black box: - Budget fit (large positive inside the verified band, heavy penalty outside) - Bedroom minimum - Lifestyle weighting: green space score if they value outdoors or have a dog; school scores if they mention schools; low-crime weighting if they want quiet - Commute to their anchor location - Value: price per square metre versus the local average from real sold prices - Energy rating (running costs and green mortgage eligibility) - A learned adjustment from swipe history once volume exists Display as a 0–100 badge with three or four plain-English reasons. Explainability is a feature, not a nicety — it is what makes recommendations trustworthy in a high-stakes purchase. ## 3.4 Property detail Photo carousel with arrows and dots; price, address, key facts; "why this scores X for you"; neighbourhood intelligence grid (schools, transport, green space, safety, broadband, amenities) with a one-line area summary; verified material information (tenure, council tax, EPC, broadband, flood risk, mining); **real sold comparables from Land Registry open data**; description; agent card with verified performance stats; live viewing slots; and the offer panel. ## 3.5 Viewings Tap a real slot from the agent's diary. Confirmation, reminders, directions, and a 30-second structured feedback prompt afterwards that also feeds the recommendation model. ## 3.6 Offers and mediation The offer carries an **evidence pack** assembled automatically: verified decision in principle, proof of deposit, source of funds, chain status, solicitor readiness. This is the product's commercial edge — a proven-proceedable buyer beats a higher offer that cannot prove itself. **Fair Negotiation — design it as a mediator, not a negotiator:** - Both sides talk privately to an independent intermediary - It never reveals either side's limits or confidences - It never advises on price; it may cite public sold-price data as context, clearly labelled - Every offer is relayed **in writing** — under the Estate Agents Act 1979 the agent must pass written offers to the seller promptly, which makes this legally effective even with no integration - The human confirms every formal step; nothing binding happens automatically - Full transcript retained; sample conversations for quality review On acceptance: generate the memorandum of sale, notify all parties, create the transaction, and open the journey. ## 3.7 Finance Verified Buying Power card. Then whole-of-market products ranked by **true monthly cost**, each showing lender, rate, fix type, fee, maximum loan-to-value, features, and whether that lender accepts applications by API. Choosing one issues an instant decision in principle; a second tap submits the full application, pre-filled from data already held. ## 3.8 Journey The single timeline: progress bar with stage label and percentage, fall-through risk, chain status, the professional team, the document pack with signing badges, and every event in reverse order with plain-English explanations and who did it. The buyer never chases anyone. ## 3.9 Documents and signing Twelve document types, generated from live transaction data: Memorandum of Sale, Client Care Letter, Property Information Pack, Search Results, Survey Summary, Decision in Principle Certificate, Mortgage Offer, Report on Title (in plain English), Contract of Sale, TR1 Transfer Deed, SDLT Return, Completion Statement. Signing: contracts by ordinary electronic signature; the **TR1 transfer deed by Qualified Electronic Signature, which requires no witness** because the signatory's identity was cryptographically verified — this is accepted by HM Land Registry and is one of the most striking things to show a customer. ## 3.10 Move-in Unlocks at exchange. Utilities, broadband, removals, insurance, cleaning — one tap each, pre-filled with the new address and completion date, every one optional. ## 3.11 Own Post-completion retention engine, near-zero marginal cost: live value estimate and equity, mortgage renewal countdown, local planning applications, sold prices nearby, improvement return estimates, document vault, and a one-tap route to selling. ## 3.12 In-app assistant A conversational assistant with access to the buyer's own case. Practical design that works well: answer the common questions instantly from templates using live case data (stamp duty with the real calculation, searches, deposit safety, timelines), and pass anything novel to a language model with the case context in the system prompt. Cache answers. Always degrade gracefully to "a human will pick this up" rather than showing an error. --- # PART 4 — THE STAFF BACK-OFFICE Every automated flow needs a human console behind it. - **Dashboard** — live counts, pipeline by stage, exception inbox (the things automation could not resolve), and a live activity feed of every API call and milestone - **Transaction Ops** — kanban board by stage; case files with the full audited event timeline, documents, mortgage status, chain, and parties - **Listings QA** — every listing checked against material information rules at ingestion. Incomplete listings are **quarantined and never shown to buyers**, with fix tasks to the agent - **Estate Agents** — partner list with feed health, verified performance data - **Mortgage Desk** — application pipeline; API lenders flow straight through, portal-only lenders surface here for a case handler, invisible to the buyer either way - **Trust & AML** — identity, source of funds and sanctions checks, with a review queue - **Revenue** — the revenue streams, showing which are live --- # PART 5 — INTEGRATIONS: THE REAL RESEARCH Everything below was researched against current UK sources. Costs are indicative. ## 5.1 Listings — the hard constraint **UK listing distribution is CRM-push. There is no central MLS, and no portal offers an outbound feed.** Rightmove, Zoopla and OnTheMarket feeds are inbound-only. Scraping them breaches their terms and carries real legal risk under consumer protection law. **Listings must come from agents, with consent. There is no shortcut, and this is the single biggest constraint on the business.** How to get fed, in order of cost-effectiveness: 1. **Publish a feed endpoint that clones the Rightmove Real Time Data Feed specification.** OnTheMarket did exactly this so that every CRM vendor's existing integration code ported across with minimal work. The spec is public. 2. **Accept the older BLM format over SFTP** — nearly every CRM can already emit it. Cheapest way to onboard the long tail of small agents. 3. **Sign feed agreements with the major CRM vendors.** These are commercial conversations, not just technical ones. 4. **Feed middleware** (around £20 per agent per month) for stragglers. **CRM APIs — the buyer-interaction surface** (viewings, offers, progression): | CRM | API | Viewings | Offers | Access | Cost | |---|---|---|---|---|---| | Reapit (enterprise) | Self-serve developer platform, OAuth2, webhooks, sandbox | Yes | Yes | App marketplace, per-agency install | Per-call, with a monthly minimum per app | | Street.co.uk | REST + webhooks + sandbox | Yes | Growing | Agency grants a token | Free | | Dezrez Rezi | JSON REST, OAuth2, webhooks | Yes | Yes | Vetted onboarding | On application | | Alto (largest SME base) | Emerging; legacy XML read | Emerging | In development | Partner deal | On application | | Apex27 | Open API + webhooks | Yes | Yes | API key | Low/free | | Jupix, 10ninety | Feed only | No | No | Per-agency | Low | **Strategic warning:** Alto is owned by the group that owns Zoopla — a competitor to any buyer-side intermediary. Have a fallback. **Viewings** can only be booked programmatically through CRM APIs. Fallback is a structured email lead. Outsourced viewing networks exist and agents already accept third parties conducting viewings. ## 5.2 Listing compliance Material information rules (formerly Parts A/B/C: price, tenure and leasehold detail, council tax, construction, utilities including broadband and mobile, parking, building safety, covenants, easements, flood risk, planning, mining) remain a legal duty. Enforcement moved to the competition regulator with **direct fines of up to 10% of global turnover** for misleading omissions. Fake or incentivised reviews are also banned — relevant if you display agent ratings. **Build the listing schema to those fields plus the Property Data Trust Framework open JSON schema**, and quarantine anything incomplete. Government reform will mandate upfront digital property packs; building to that standard now is a two-to-three-year head start. ## 5.3 Mortgages **Regulatory gate:** ranking or recommending products, or submitting applications, is a regulated activity. No authorisation means no product ranking, no decision in principle, no submission. A 2025 rule change removed the requirement that any interactive dialogue forces an advised sale, which makes a well-designed execution-only digital journey viable for the first time — but presenting a "best deal for you" is still advice. Route: launch as an **appointed representative** of an authorised network (weeks, roughly £300–£1,000 a month) or as an introducer to a partner broker (days), while applying for **direct authorisation** in parallel (6–12 months, application fee plus compliance consultants, minimum capital, qualified advisers, professional indemnity insurance). **Sourcing and submission platforms:** the major sourcing engines offer APIs for product sourcing, criteria and affordability; one lender-owned consortium platform provides a multi-lender decision-in-principle and full-application gateway **free to intermediaries, with an API for third-party systems** — the most relevant rail. Modern broker platforms offer one-click decisions with specific major lenders. **Reality check:** roughly 8–10 large lenders are reachable by API, covering a large share of volume. The rest remain broker-portal-only. "Whole of market, fully automated" is not achievable today — build a hybrid: API where available, a human desk everywhere else, with an identical experience for the customer. **Verification:** open banking providers for income and expenditure (you must be a registered account information provider or act as an agent of one — the agent route is fastest); credit bureaux require vetted agreements, so start via resellers; there is no third-party API for HMRC tax calculations, so self-employed income is either user-supplied or verified through a consent-based employment verification provider. **Handoff:** lenders instruct panel conveyancers through panel-management platforms. The Certificate of Title triggers the release of funds, typically with 3–5 working days' notice. An app can orchestrate and track this, but cannot execute it. ## 5.4 Conveyancing **Hard legal boundary:** preparing transfer deeds and lodging registration applications are reserved legal activities. Doing them unauthorised is a criminal offence. Everything *around* that core — identity checks, ordering searches, gathering information, drafting enquiries, chasing, form pre-population, progress tracking — is unregulated and fully automatable. That is roughly 80–90% of the work. Route: start with **white-label partner firms** (fast, no regulatory burden, margin shared), then acquire or licence a regulated practice at volume. Acquiring an existing firm also buys its **lender panel memberships**, which a brand-new firm cannot quickly obtain — this is the hidden moat, and it catches people out. **Land Registry APIs** (business gateway): official copies of the register and title plan, official searches with priority, bankruptcy searches, an instant local land charges API where the council has migrated, and electronic document registration. Access requires a business account with a credit facility; **registration services are restricted to authorised conveyancers** — so the software integrates as the firm's system, not as an independent actor. **Electronic signatures:** qualified electronic signatures are accepted for transfers, charges and assents **with no witness required**, because the trust provider's identity check replaces the witness. Ordinary platform signatures still need a physically present witness. Remote video witnessing is not valid. Contracts (as opposed to deeds) need no witness at all. **Searches:** local land charges can be instant where digitised, but the local authority search is still council-fulfilled and varies from days to six weeks — this is the main uncontrollable delay. Aggregator APIs cover all search types through one integration with webhook callbacks. **Upfront packs:** the open Property Data Trust Framework schema plus the standard upfront property information dataset. A major industry pilot demonstrated a **35% reduction in the time from sale agreed to exchange** using upfront packs. Note that the standard property forms are copyright — either licence them or generate equivalents from your own data. **Completion:** funds can flow through the conveyancer's client account or a regulated third-party managed account. Digital settlement networks now handle remortgage and, more recently, sale and purchase — ride those rails rather than building payment plumbing. **Stamp duty** returns can be submitted by API. Note that paid filers must now register with the tax authority as agents. **Benchmarks:** tech-first conveyancers achieve 5-week guaranteed exchange on freeholds (hit about 80% of the time) against a 10–16 week market average. ## 5.5 Identity, anti-money-laundering and payments **Identity to the Land Registry digital standard** requires reading the chip in a biometric passport cryptographically, plus a liveness check matched to the chip photo; sellers also need a check connecting them to the property. Meeting this standard gives the conveyancer a safe harbour against fraud claims. Use a certified provider — several specialise in property, with per-check costs from a few pounds to around £20 depending on depth. Source-of-funds checks via open banking are a separate, specialist product. **Anti-money-laundering:** if the platform introduces buyers to sellers or handles offers, it is doing estate agency work and **must register with the tax authority for AML supervision before trading — operating unregistered is a criminal offence.** Registration is cheap; the operational load (a nominated officer, risk assessment, checks on both parties, ongoing monitoring, suspicious activity reporting, five-year records) is the real cost. Also required: membership of an approved redress scheme, and professional indemnity insurance. Note that formal reliance on another firm's checks leaves liability with the relying firm, so in practice you share the *evidence* with consent rather than promising conveyancers you have done their compliance for them. **Payments:** never hold purchase funds on the platform's balance sheet — that is deposit-taking and requires a banking licence. Funds flow through the conveyancer, or later through a regulated third-party managed account. Ordinary card processing is fine for fees. **Fraud — build this in, do not bolt it on.** Payment diversion fraud costs UK buyers an average of about £78,000 per incident. Bank details must never travel by email: keep a payee vault inside the platform, verify every payee against the account-name-checking service, freeze on any change of details with out-of-band re-verification, and require multi-factor authentication on any screen showing payment details. ## 5.6 Property data **Use the free government data — it is excellent and commercially licensed.** Key the whole system on the Unique Property Reference Number. - **Land Registry Price Paid** — every sale since 1995, monthly updates, free, commercial use permitted with attribution. This powers real sold comparables and is one of the most credible things you can show a buyer. - **Energy performance certificates** — floor area, rooms, age band, heating. Floor area is the crucial free input for any price-per-square-metre valuation. - **Flood risk, planning designations, crime, schools and inspection outcomes, census and deprivation data, green space, address/UPRN reference data** — all free. - **Broadband and mobile coverage** — free API from the communications regulator. **Commercial data:** a self-serve property data API covering valuations, comparables, yields and planning starts at around £28–£300 a month depending on volume — the sensible starting point. Lender-grade automated valuation models exist at enterprise pricing when credibility matters. **Building your own valuation model** from sold prices plus energy certificate floor areas and locality features is genuinely viable for a first version — expect wider error margins than a commercial model, so always display a range with a confidence band and describe it as a guide, never a valuation. **Lifestyle scoring inputs:** transport APIs, open street map data self-hosted for amenities (note the share-alike licence on derived databases), air quality, green space. Mapping platform place data generally may not be cached or used to build stored scores — check the licence before designing around it. **Surveys:** no public booking API exists. Route is a referral partnership with a digital-first survey network or a panel deal, typically 10–20% commission. **Move-in services:** home-setup providers offer genuine partner APIs for utilities, council tax and broadband; switching platforms offer white-label journeys; insurance requires an introducer or appointed representative arrangement. All commission-based, and the easiest early revenue in the whole model. --- # PART 6 — REGULATORY ROADMAP The order matters. Trading before some of these is a criminal offence, not a compliance risk. **Before any buyer-side activity:** 1. Incorporate; consider separate entities so that regulated activities are ring-fenced 2. **Approved redress scheme membership** 3. **AML registration with the tax authority** — before trading 4. Professional indemnity insurance; appoint a nominated officer; write the risk assessment **Mortgages:** appointed representative or introducer to launch; direct authorisation applied for in parallel. **Conveyancing:** white-label partner firms first; own or acquire a regulated practice at volume. **Money:** never held by the platform. **Direction of travel is favourable.** Government reform will mandate upfront digital property packs, identity verified once and reused across parties, and binding commitments earlier in the process. Legislation lands over the next few years. Building to those standards now is a material head start. --- # PART 7 — WHAT THE MARKET TEACHES Read this before deciding strategy; it is the most expensive knowledge in this document because other people paid for it. - **Every surviving end-to-end player monetises the transaction, not discovery.** The clearest current example grew revenue several hundred percent by taking mortgage and legal margin while listing was free to agents. - **Portal challengers that tried to out-feature the incumbent died.** One well-funded challenger launched free, flipped to charging agents during a market downturn, failed to raise, and liquidated within about twenty months. You cannot beat the dominant portal at classified listings. - **Never ask agents for exclusivity.** An earlier challenger's rule restricting which other portals members could use bred lawsuits and resentment. - **Agents' marketing budgets are exhausted.** Portal fees already consume a large share of agency commission and have risen well above inflation, prompting a collective legal claim. That resentment is your opening — but it means "add us, we're free" works and "replace them" does not. - **Consumer-brand acquisition costs without transaction capture is fatal.** A well-known digital mortgage broker only became viable as a feature inside a bank's existing app. - **Listing-fee models that decouple payment from completion fail.** Paying to list rather than to sell misaligns everyone. - **Swipe interfaces alone are not a moat.** A "Tinder for property" app entered liquidation owing millions. Swiping is a good preference-capture device and a great demo; the defensible layer is what happens after the shortlist. The traction in 2026 is with conversational, AI-native search. - **AI mediation in residential property is genuinely unoccupied.** But doing it makes you an estate agency business in law, with the redress, AML and insurance obligations that follow. A regulator has already authorised an AI-first law firm on the basis that no autonomous step happens without client approval and named humans stay accountable — a good template. --- # PART 8 — SUGGESTED BUILD ORDER **Phase 0 — Foundations.** Redress scheme and AML registration filed. Mortgage introducer agreement signed. Two white-label conveyancing firms contracted. Then: data model, event-sourced transaction, property data ingestion pipeline keyed on UPRN, listing schema built to the compliance standard, feed endpoint published, and the swipe deck with explainable scoring. **Phase 1 — Pilot city.** One city, done properly. Agent onboarding, viewing booking, buyer verification, the qualified-offer mechanism, the journey timeline over partner conveyancing, and the staff back-office. Monetise from day one through conveyancing, survey, mortgage and move-in referrals. **Phase 2 — Own the mortgage, deepen the legal.** Appointed representative live, direct authorisation filed, sourcing and submission APIs integrated, qualified electronic signing live, upfront property packs generated, chain visibility integrated, ownership dashboard on. **Phase 3 — Own the rails.** Direct authorisation granted, regulated conveyancing capacity owned or acquired, digital settlement network membership, assisted selling, and the data and intelligence products. **Deliberately do not do early:** own conveyancing firm before volume; mortgage advice before authorisation; holding client money; native apps before product-market fit. --- # PART 9 — HONEST RISK REGISTER **Structural, cannot be engineered around:** 1. No portal-side listings — supply must come from agents, agency by agency. **No agent consent, no inventory. This is existential.** 2. Regulated capacity gates every "cut out the middleman" promise. 3. Lender API coverage is partial; a human desk is required behind the scenes. 4. A new law firm cannot easily get onto lender panels. 5. Local authority searches remain council-speed. 6. Chains are the uncontrollable variable. Never promise chain-proof timelines. **Business risks:** the supply-side cold start; incumbents converging on the same vision with distribution you do not have; service quality of volume conveyancers becoming your brand problem; and the temptation to monetise agents later, which has killed challengers before. **Compliance risks:** material information duties with turnover-based fines; AI mediation outputs carrying the same legal duties as a human agent's; payment-diversion fraud; and data licensing hygiene across the free datasets. --- ## Closing note Everything here is offered freely and without conditions. It reflects genuine research into UK property technology, the regulatory position, and the practical realities of getting a transaction platform live — plus the lessons of a working prototype that ran the whole journey end to end. The idea is good. The market timing is unusually favourable. The hard part is not the software — it is the agent relationships and the regulatory sequencing. Spend your energy there. Good luck.