Offrly Sale Valuation Accuracy Report (2026)
Offrly sale valuations land within ±10% of the price the same home sold for in 68.2% of cases — across 22 homes matched to their HM Land Registry sale. Misses included.
Every BUY valuation is sealed at quarter-end — postcode, house number, prediction, timestamp.
Latest valuation per home only — nothing counts twice.
Same postcode, same house number or name: that home's own Land Registry record, never a neighbour's.
Its open-market sale closest in time, before or after, within two years. Never the closest in price.
Hit rates, MdAPE, MAE — full distribution, misses included.
How one pair resolves
IllustrativeDeterministic, dated, traceable to a real prediction and the real HMLR record of the same home. No human curation anywhere in the match. The numbers here are made up for illustration — no specific home is implied.
By property type
Each type needs ≥ 30 homes to reportA 68.2% hit rate means roughly 7 in 10 estimates landed within ±10% of the real sold price — compare that with the 60–70% mainstream postcode-average AVMs typically report. The difference is architecture: photo-aware comparables and street-level pricing mean two very different homes on one road get two very different numbers. It is still a market estimate, not a guaranteed sale price — for probate, mortgage or insurance, use a RICS surveyor.
This page pools every quarter. The benchmark shows each snapshot alone — dated, immutable.
Open the quarterly benchmark →How we measure accuracy
Offrly's accuracy figures are computed against HM Land Registry Price Paid Data — the authoritative record of every residential property transaction in England and Wales, published monthly under the Open Government Licence.
Each figure scores a real Offrly prediction against the price the same home sold for — not a neighbour's sale on the same postcode, and not a re-run over the data the model learned from. Every pair traces back to a real HMLR sold-price row, with no admin in the loop:
- Every BUY valuation Offrly issues in a given quarter is frozen at quarter-end into an immutable snapshot — postcode, house number or name (where one was given), predicted price, timestamp and the confidence band shown with the estimate (a 90% interval since July 2026; earlier valuations displayed a 95% band).
- Across every quarterly snapshot we collapse to the latest BUY valuation per home. A home revalued in a later quarter only contributes its most recent prediction, so no home lands in the sample twice.
- Each valuation is matched to that home's own HM Land Registry record: the same full postcode and the same house number or name (for a flat, the flat number). A sale of another home on the same postcode never counts. Valuations without a house number can't be tied to one home, so they aren't scored — but they stay in the denominator below.
- Of that home's open-market sales, we take the one closest in time to the valuation — before or after it, within two years — never the one closest in price, which would flatter the model. Repossessions and other non-market transfers are left out, and so is any valuation whose property type doesn't match the Land Registry record: it described a different home.
- The metrics on this page aggregate every quarterly snapshot into a single lifetime view: hit rate within ±10% and ±20%, MdAPE and MAE. The same matching runs on the per-quarter benchmark.
What's in the report, transparently
- Scope: sale valuations only. Rentals are not included — there's no public sold-price equivalent for rentals to pair them against.
- Ground truth: HM Land Registry Price Paid Data — every arms-length freehold and most leasehold residential sales in England and Wales.
- Licence: Open Government Licence v3.0 — reusable, citable.
- Sample: 22 homes matched to their own HM Land Registry sale (0 sold after our valuation, 22 before it) — out of 361 homes valued with a house number, and 1,387 sale valuations frozen since 2026-04-26 (1.59% matched). We deliberately publish the base-rate denominator so coverage is auditable.
- Leakage controls: an Offrly valuation never looks up the Land Registry record of the home being valued, so a sale that had already completed can't leak into its estimate. A sale after the valuation is a true forecast; the sample line above says how many of each there are.
- Update cadence: a new snapshot freezes when each quarter closes, and matches are added as HM Land Registry publishes each month's sales. The page recomputes whenever a snapshot or the Land Registry data changes.
How we compare to the rest of the market
Mainstream automated valuation models — the postcode-average tools most free portals run on — publish hit-rate-within-±10% figures broadly in the 60–70% band. Offrly's differentiator isn't just the point estimate, it's the model architecture: a photo-aware AI that reads each comparable the way a human analyst would, paired with hyperlocal pricing that resolves down to the street rather than the postcode. The price bracket is free, with no sign-up — and the methodology is fully auditable on this page.
All accuracy claims age quickly as the market moves. As fresh HMLR data lands and new quarterly snapshots resolve, the figures here update.
For the per-quarter view of the same data, see the sealed quarterly benchmark — each individual snapshot, dated and immutable. Both pages compute their metrics from the same matching; this one just pools every quarter so the sample is bigger.
Frequently asked questions
What does 'hit rate within ±10%' mean?
It's the share of Offrly sale estimates that land within 10% of the actual sold price across every valuation we've matched to an HM Land Registry sale. A 60% hit rate means 6 in 10 estimates were within ±10% — an industry-standard way of measuring how close an automated valuation model gets to the true price.
How are predictions paired to HMLR sales?
Every BUY valuation issued in a given quarter is frozen to an immutable snapshot at quarter-end. Each one is matched to the HM Land Registry record of the same home: the same full postcode and the same house number or name (for a flat, the flat number). A sale of another home on the same postcode never counts. Each home counts once, with its latest valuation, paired with that home's open-market sale closest in time to the valuation, before or after it, within two years. Valuations without a house number can't be tied to one home, so they aren't scored. There is no human review.
Why do sales from before the valuation count?
Because an Offrly valuation never looks up the Land Registry record of the home being valued, so a sale that had already completed can't leak into its estimate. Counting only sales after the valuation would leave the report nearly empty: few homes sell within a year of being valued. We show how many of the matched homes sold after the valuation and how many before, and the two-year limit stops market movement passing for model error.
Does this cover rentals?
No. The accuracy report is for sale valuations only. There is no public sold-price equivalent for rentals, so we cannot publish a comparable rental accuracy figure honestly.
How is this different from the per-quarter benchmark?
The metrics on this page aggregate every quarterly snapshot into a single lifetime view. The /benchmark URL shows a single quarter's snapshot, sealed and dated. Both pages compute their metrics with the same matching — this one just pools the snapshots so the sample is bigger.
Why should I trust these figures more than a marketing claim?
Because the ground truth is HM Land Registry, published under the Open Government Licence — and the pairing is fully automated, with no admin choosing which pairs make the cut. Each figure on this page traces back to a frozen Offrly prediction and the HM Land Registry sale of the same home — never the closest-priced sale nearby, which would flatter it.
Should I use an Offrly valuation instead of a surveyor?
No. Offrly estimates are indicative market guidance. For probate, mortgage, insurance or tax-binding figures, use a RICS-qualified surveyor. Offrly is for orientation, negotiation and informal decision-making — not for binding purposes.
Source: HM Land Registry Price Paid Data, licensed under Open Government Licence v3.0. Offrly estimates are indicative market guidance — not regulated valuations and not financial, tax or legal advice.