Case Study · Appraised

A property-valuation engine built on public data alone

Appraised values any home in England and Wales in seconds, from public records only. No paid valuation feed sits behind it. We built the engine, the two-sided platform around it, and the introducer model that earns a fee when a sale completes.

Public data

The only inputs. No licensed AVM.

Seconds

From postcode to a grounded estimate

England & Wales

Every address the Land Registry covers

200+

Market pages from the same engine

The problem

The hard part was never the website

Appraised began from a model that already worked: introduce a homeowner to the right local agent, take a fee when the sale goes through. Proven over years, but run on spreadsheets and inboxes. Turning that into software was the straightforward half.

The valuation was the hard half. To pull homeowners in, every estimate had to be trustworthy and free, for any address in the country. The usual answer is to licence an automated valuation feed and move on. We didn't want one. A bought number is a black box you can't explain to a seller, and a bill that grows with every search. So the real brief was narrower and much harder: value any home in England and Wales from public data alone, and make it cheaper over time, not dearer.

Public data only

Land Registry, EPC and HPI. Nothing licensed.

Free at the point of use

Cost per valuation had to fall close to zero.

Every address

National from day one, not three pilot cities.

The solution

One system, with the engine at the centre

We rebuilt Appraised as a single platform with the valuation engine at its core. Three decisions did most of the work.

01 · Two-sided by design

Homeowners on one side, agents on the other

Homeowners get an instant estimate and invite local agents to pitch. Agents get leads they only pay for on completion. The platform introduces the two sides and then gets out of the way.

The homeowner

  1. 1Get an instant estimate
  2. 2Invite local agents to pitch
  3. 3Compare proposals side by side
  4. 4Pick an agent, or sit tight

The agent

  1. 1Receive the opportunity by email
  2. 2Send a valuation, fee and plan
  3. 3Get compared on merit
  4. 4Win the instruction; pay on completion

Both journeys run off one property record. The platform makes the introduction and stays out of the way.

02 · One record per property

A single source of truth

Every valuation, agent proposal, comparable and market snapshot hangs off one property record behind row-level security. The seller, the agents and the office all read from the same place.

03 · Reply from your inbox

The piece I'd show first

An agent can answer a valuation request straight from email, with no account and no login. A signed token carries the property and the permissions. It is the difference between a reply in two minutes and no reply at all.

The algorithm

Valuing a house you've never seen

This is the part that earns the hire. No site visit, no licensed feed. Just public records and a method you can defend line by line.

Land Registry

Real sold prices

EPC Register

Floor area & type

UK HPI

Inflation adjustment

Live market

Asking prices (cap only)

The engine

Widen the search in rings until there are enough real comparables.

Price per square metre where EPC data allows; price-based where it's sparse.

Asking prices can pull the figure down, never up.

The estimate

  • A figure backed by real sales
  • A confidence range, not false precision
  • The comparables behind it
  • A cache that sharpens the next one

Widening the search

01Streetimmediate neighbours
02SectorWN5 1
03DistrictWN5
04AreaWN

The engine stops at the first ring with enough genuine comparables. A dense terrace resolves on its own street; a rural address has to reach wider.

Every estimate starts from ground truth: the sold prices the Land Registry holds for England and Wales, enriched with floor area and property type from the EPC register. From there the engine widens its search in rings, from the street out to the wider area, stopping the moment it has enough genuine comparables. A terraced house gets valued against its terrace. A farmhouse reaches further, because it has to.

Where EPC floor area exists, it values by price per square metre. Where the data is thin, it falls back to a price-based model, with the UK House Price Index dragging older sales up to today's money so a 2019 comparable still counts.

Then the reality check, and this is the bit I like. Live asking prices can pull an estimate down but never up. Asking prices are hope, not evidence: a hopeful neighbour shouldn't inflate your number, but a genuinely cheap nearby sale should temper it. Every figure ships with a confidence range rather than false precision.

The last decision is the one a bought feed can never match. Every lookup is cached, so the dataset gets richer each time someone uses it. The valuations sharpen and the marginal cost falls. We didn't licence an asset. We built one that compounds.

The stack

Boring where it should be

A deliberately unremarkable stack, so the interesting engineering could live in the engine and the data, not the plumbing.

Next.jsReactTypeScriptTailwind CSSSupabasePostgreSQLDocRaptorResendgetAddress.ioGoogle MapsLand Registry PPDEPC RegisterUK HPI

Got a model that works, but tools that don't?

Let's build the system it deserves.

Start your project