How Offrly Works
An honest, outcome-oriented view of what Offrly produces, the pipeline behind every estimate, where it's reliable, and how we measure accuracy in public.
The short version
Offrly is a free UK property analyst with two tools — AI-graded property search and instant sale or rental valuation. Both run in about 30 seconds, free, with no mandatory signup. The estimate is indicative market guidance — not a regulated valuation, not financial / tax / legal advice.
Below: what we produce, what's in the estimate, what's not, where it's reliable, where it isn't, and how we publish accuracy.
What you get from a valuation
- A point estimate of sale price (or monthly rent if you switch to the rent channel).
- A fit score (R²) showing how well the surrounding sales explain price variation in this specific local market — see What R² means here below.
- The top 100 comparables (out of typically 300–500 evaluated for a query), so you can sanity-check the answer yourself.
- A per-property-type signal — flats are matched against flats, leasehold against leasehold, bungalows against bungalows.
What you don't get: a regulated valuation, a building-condition survey, a financial-advice or tax answer.
What you get from a property search
- A natural-language brief — describe what you want in plain English.
- Live UK listings pulled at query time (sale or rent).
- Each listing graded 0–100 for how well it matches your brief.
- Reasoning tags that explain each score so you can see why a listing scored.
- An iteration loop: tell the AI what you want more or less of and the shortlist re-ranks.
What you don't get: a viewing booking, a mortgage decision, an offer process. Offrly is intelligence; the transaction layer is whoever is selling or letting the home.
How we get to a price
Every Offrly valuation runs the same five-stage pipeline. There's no human in the loop.
- Find — pull comparable sales and live listings around the target postcode within a search radius. Typically 300–500 comparables make it through this step.
- Clean — drop plots and land-only entries, drop price outliers, and drop comparables whose photos couldn't be condition-rated. The AI declines to guess on signal it can't see.
- Read — for each surviving comparable, capture distance from the target, beds and baths, property type, tenure, plus what photo-aware AI gleans from listing photos the way a seasoned property analyst would.
- Fit — fit a price model on the cleaned, scored comparables. The model weights each feature based on how well it explains observed prices in this specific local market, which is why R² varies between queries.
- Apply — plug the target's features into the fitted model to get the estimate, and report R² as the fit quality.
Worked example
A 3-bed mid-terrace in a typical inner-London postcode. The pipeline evaluated 380 comparables within scraping range. Below are the top 10 by weight — the comparables the pricing model leaned on most:
| # | Comparable | Asking / sold | Beds/Baths | Distance | Garden | Condition | Tenure | Why it counted |
|---|---|---|---|---|---|---|---|---|
| 1 | Same road, 4 doors down | £640,000 | 3 / 1 | 0.04 mi | Good | Good | Freehold | Closest tenure-and-type match |
| 2 | Adjacent road, mid-terrace | £625,000 | 3 / 1 | 0.10 mi | Avg | Good | Freehold | Strong type match, close |
| 3 | Same postcode, end-terrace | £680,000 | 3 / 2 | 0.18 mi | Good | Good | Freehold | End-terrace premium |
| 4 | Same postcode, mid-terrace | £590,000 | 3 / 1 | 0.22 mi | Avg | Dated | Freehold | Condition discount |
| 5 | Adjacent ward, mid-terrace | £655,000 | 3 / 1 | 0.31 mi | Good | Excellent | Freehold | Renovation premium |
| 6 | Same road, 2 doors down | £600,000 | 3 / 1 | 0.02 mi | Avg | Dated | Freehold | Closest, down-weighted on condition |
| 7 | Adjacent road, mid-terrace | £620,000 | 3 / 1 | 0.14 mi | Avg | Good | Freehold | Type & tenure match |
| 8 | Same ward, mid-terrace | £645,000 | 3 / 1 | 0.36 mi | Good | Good | Freehold | Borderline radius, light weight on distance |
| 9 | Same postcode, semi | £710,000 | 3 / 2 | 0.24 mi | Large | Good | Freehold | Type penalty (semi vs terrace) |
| 10 | Adjacent ward, mid-terrace | £635,000 | 3 / 1 | 0.41 mi | Avg | Good | Freehold | Furthest in the top 10, lightest weight |
Estimate: £625,000. Fit score (R²): 0.87 — the surrounding sales explain 87% of the local price variation, so we're in a tight market.
The model up-weights tenure-matched, type-matched, photo-rich comparables that are physically close. It down-weights comparables that are further away, in a different sub-type, or where photos suggest substantially different garden / condition. Comparables 1, 2 and 7 carry the most signal here; comparable 9 (a semi) gets a type penalty even though it's nearby.
What R² means here
R² (the fit score) is how well the surrounding sales explain price variation in this specific market.
- 0.85 or above — tight, well-comped market. Estimate is reliable.
- 0.65 to 0.85 — typical market. Estimate is a sound starting point; verify outliers manually.
- 0.50 to 0.65 — heterogeneous market. Treat the figure loosely.
- Below 0.50 — thin or unusual market. Read the estimate as a rough anchor, not a number to act on.
R² is reported on every result card. It is also the input we use ourselves when deciding how much weight to give the model on a per-query basis.
What's IN the estimate
- Comparable evidence. Live UK listings nearby, pulled at query time. Nothing cached from last month.
- Photo signals. The AI reads each comparable's listing photos for garden quality, internal condition, layout, finish — the same signals a seasoned property analyst would weigh during a viewing.
- Hyperlocal pricing. Prices are resolved closer to the street than to the broad postcode area. Two homes on the same road, one immaculate and one tired, get materially different numbers.
- Tenure differentiation. Leasehold, share-of-freehold and freehold homes are matched against tenure-equivalent comparables. We don't borrow freehold prices to value a leasehold flat.
- Property-type differentiation. Bungalows are valued against bungalows; detached homes against detached. Each segment trades on its own demand profile.
What's NOT in the estimate (and matters separately)
These items genuinely affect what a UK home is worth on the open market — but they don't reliably show up in listing photos or comparable sale prices, so the AI does not see them. Treat them as buyer-side due diligence items, not failures of the estimate:
- Lease length (years remaining on the lease). Material below 80 years; can impair mortgage availability below 70. The number is in the seller's pack — your solicitor will pull it.
- Ground rent terms, especially escalation clauses (10-year doubling, RPI-linked, fixed-step).
- Service charge and sinking-fund contributions, plus any pending major-works estimates.
- Cladding / EWS1 status in higher-risk buildings — affects mortgage availability and price.
- Specific structural / planning issues — subsidence, restrictive covenants, planning enforcement, party-wall disputes, drainage / off-grid utility concerns.
- Listed-building obligations and conservation-area planning constraints that affect what you can do with the property.
- Boundary disputes, easements, and any access-right restrictions.
The right tools for those: the seller's pack, a RICS-qualified surveyor, and your solicitor. Use Offrly's estimate as a fast, free first-pass market answer; use those professionals to close the loop.
Where Offrly is most reliable
- Standard freehold homes (detached, semi-detached, terraced) in well-populated areas with active recent comparable activity.
- Standard leasehold flats where the lease is a normal term (typically 99/125/250 years, low ground rent, ordinary service charge) — though we still recommend verifying lease finance separately.
- Postcodes with enough live comparables to populate the model with real signal.
- Photo-rich listings — when comparables include lots of photos, the AI has more to work with.
Where Offrly is less reliable
- Unusual stock. Listed buildings, converted commercial, mixed-use, exceptionally-large estates.
- Very short leases (typically under 70–80 years remaining). The market prices these on lease-extension premium calculations the model does not have access to.
- Brand-new schemes with no prior resale history. The model leans on similar-spec resales nearby, which is a less direct signal.
- Very rural / unique properties with few comparable listings within scraping range.
- Photo-poor listings — where there's little to read, the AI's added value over a postcode-average tool shrinks.
When confidence is low, the R² fit score will be low — typically below 0.65. Read the estimate loosely in that case.
Coverage
Offrly covers the United Kingdom only — England, Scotland, Wales and Northern Ireland — subject to live comparable listings being available in the area. We do not cover Ireland, the Channel Islands, the Isle of Man or any non-UK market.
For sold-price intelligence on location pages, England and Wales coverage comes from HM Land Registry Price Paid Data under Open Government Licence v3.0. Scotland and Northern Ireland have separate registers; we use city-level summaries for those.
How we measure accuracy in public
Two transparent views over the same auto-joined pair set, sitting alongside each other:
- Accuracy report — lifetime aggregate. Every BUY valuation Offrly has frozen across every quarterly snapshot, auto-joined to HM Land Registry sales on the same full postcode (closest-by-price tie-break, documented openly). Largest pooled sample.
- Benchmark — sealed per-quarter snapshots. Each calendar quarter's snapshot is dated, immutable, and published as a single page so the metrics for that quarter cannot be re-curated, re-cherry-picked or re-tuned after the fact.
Both are prediction-vs-outcome reports, not synthetic back-tests, with no human in the loop. Each figure traces back to a real Offrly prediction made strictly before the eventual sold date — predictions never see future sold prices.
Rentals are not in either report — there is no public sold-rent register equivalent to pair predictions against.
What Offrly does not do
- It is not a regulated valuation. For mortgage, insurance, probate, tax-binding or court purposes, you need a RICS-qualified surveyor.
- It is not financial, tax or legal advice. For those questions, see the relevant qualified professional.
- It is not a long-term price forecast. Offrly produces a current-market estimate based on current comparable evidence; market conditions change.
- It is not a substitute for a building survey — Offrly cannot see structural defects, damp, subsidence, roof condition or anything else a surveyor would catch in person.
- It is not a substitute for due diligence — read the seller's pack, ask your solicitor about lease finance and search results, and view the property.
Feedback
If you think an Offrly estimate is materially off, let us know via the contact form. Off-by-a-lot examples — with the address or postcode and the headline reason — are the most useful kind of feedback.