What is an AVM deal-score, and can you trust it?

An AVM deal-score is a single number that compares a property's asking price against an estimate of its fair market value. The estimate comes from an Automated Valuation Model (AVM) — software that values a home from sold-price data and property characteristics rather than a person visiting it. If a flat is on at £300,000 and the model estimates fair value at £330,000, the deal-score flags it as roughly 9% below estimate. That's the whole idea: estimate versus asking, expressed as a percentage or a rating.

The crucial word is estimate. An AVM is not a valuation, a survey, or a mortgage offer. It's a statistical best-guess produced in milliseconds, and it can be wrong — sometimes by a little, occasionally by a lot. A good deal-score is a triage tool that helps you decide which listings deserve a closer look, not a verdict on whether a home is worth buying.

Nestraq runs an AVM on every new UK listing and price change, then turns the result into a deal-score so you can see, at a glance, whether the asking price looks keen or rich relative to the market — with the comparable evidence shown underneath, never hidden behind a number.

AVMs are built on the same logic an estate agent or surveyor uses — comparable sales — but applied by software at scale. In the UK the backbone is sold-price data: HM Land Registry Price Paid Data for England and Wales, and Registers of Scotland north of the border. The model also pulls property attributes (type, size, bedrooms, age, tenure) and location data (postcode, transport, local price trends).

The process, simplified, runs like this: identify the subject property; gather recent sales of similar nearby homes; strip out odd transactions (a probate sale, a transfer between family); time-adjust older sales to today using a local price index; adjust for differences in size and features; then blend the adjusted comparables into one estimate, weighting the closest, most recent and most similar sales most heavily. The output is a single figure — say £330,000 — that represents the model's best read of fair value on that date.

This is exactly how the big UK AVMs behave. Zoopla's estimate is produced by Hometrack (the country's largest automated residential valuer) using Land Registry, Registers of Scotland and Ordnance Survey data refreshed monthly; Rightmove runs its own AVM across roughly 400,000 properties a month, each with a confidence score. The mechanics are well established — the art is in the data quality and how the comparables are chosen.

Every serious AVM attaches a confidence measure to its estimate, and RICS guidance actually requires it: an AVM result should come with "a measure of confidence in the accuracy of the result." Confidence might be a score (say 0–100), a band (High / Medium / Low), or an implied range (£320k–£345k). It tells you how much faith to put in the headline figure for that specific property.

Confidence rises when there are many recent, genuinely comparable sales close by, the property is a standard type for the area, and the data is complete. It falls when comparables are few, distant or mixed in type, the home is unusual, or records are patchy. This is why mortgage lenders lean on confidence so heavily. A common lender pattern: high confidence (often above ~80%) can support an AVM-only valuation at modest loan-to-value; moderate confidence (~60–80%) triggers a surveyor-led desktop valuation; low confidence (below ~60%) sends a human out to inspect.

The practical lesson for a buyer is simple: a deal-score on a high-confidence estimate is worth acting on; the same headline discount on a low-confidence estimate is mostly noise. A good tool shows you both. Nestraq surfaces the confidence alongside the deal-score, so a "15% below estimate" flag on a unique period house reads very differently from the same flag on a standard semi with ten clean comparables down the road.

AVMs are genuinely good at the bread-and-butter of the UK market: standard houses and flats in dense areas with plenty of recent comparable sales. UK analysis suggests standard residential AVMs can land within roughly ±5% in over 80% of cases, and one UK provider reports a median absolute error of about 2.8% with 90% of estimates within ±10%. For a typical three-bed semi in a busy suburb, that's a useful, fast read on value.

They struggle exactly where comparables run thin or the property breaks the mould. Period and one-off homes, heavily renovated properties, rural houses on large or unusual plots, and low-transaction postcodes all push error up — into the 15–30% range, and occasionally further. Two houses on the same street can differ by 20% on condition and refurbishment alone, and an AVM often can't see inside. Leasehold quirks matter too: if lease length, ground rent or service charges aren't well captured, the estimate can drift.

There's also a timing trap. Land Registry Price Paid Data typically lags completion by one to three months (and a small share by six months or more), so the very latest sales may not have fed the model yet. In a fast-moving local market, an AVM can be working from slightly stale evidence. None of this makes AVMs useless — it just defines where to lean on them and where to verify.

Treat the deal-score as the start of a question, not the answer. A big "below estimate" flag should prompt you to ask why. Sometimes it's a genuine motivated seller, a probate sale, or a mispriced listing — the deals worth chasing. Other times the asking price is low because the home needs work the AVM can't see, the lease is short, or it backs onto a railway line. The score points you at the listing; you do the thinking.

A sensible routine: check the confidence first, then look at the comparables the estimate is built on (are they really like-for-like and recent?), then deep-link to the source portal to study the photos, floorplan and description for the condition and quirks the model misses. Cross-check against current asking prices nearby, not just sold prices, because asking prices tell you what's happening in the market right now while sold data tells you what happened months ago.

Used this way, a deal-score does the one job it's good at: cutting a flood of listings down to a handful worth your attention. That's how Nestraq is designed to be used — continuous monitoring of new listings and price drops, an honest deal-score with the analysis attached, and a hard gate on your own must-haves, so an alert only fires when a listing is both a plausible deal and an actual fit. The judgement, and the offer, stay with you.

It's worth being clear about the line a deal-score must not cross. It is not a mortgage valuation, which the lender carries out for their own security and which can itself be just an AVM or desktop check. It is not a RICS Home Survey (Level 2 or 3), the buyer-commissioned inspection that finds damp, structural issues and repair costs. And it is not a formal RICS "Red Book" valuation you could rely on for tax, probate or a dispute.

An AVM can't walk the property, lift a floorboard, read the lease, or judge whether the kitchen extension was done properly. So even a high-confidence, deeply-discounted deal-score should funnel you towards the real due diligence: a viewing, a survey on anything older or non-standard, a solicitor's review of the lease and searches, and your own read on the area. The deal-score earns its keep by getting you to the right doorsteps faster — not by replacing the checks that protect you once you're there.

Honest framing matters here: any AVM figure, Nestraq's included, is an estimate, not a valuation, and we say so plainly. The value of a deal-score isn't precision to the pound; it's direction at scale — finding the few listings, out of thousands, that are worth your time, then handing the decision back to you with the evidence in plain sight.