What is an automated valuation model: software that estimates a property's value from recorded sales, property characteristics and market data, with no human valuer and no visit to the property
An automated valuation model, usually shortened to AVM, is a statistical model that produces a property value estimate in seconds. It takes what is on record about the property (type, size, age, location, previous sales) and what the market around it has done (comparable sales, listings, price indices), runs them through its algorithms, and returns a value with a confidence measure. No valuer forms an opinion and nobody looks at the property.
It matters because it makes valuation nearly free and instant at any scale. A lender can value an entire mortgage book overnight, screen a remortgage application in the time the page loads, or sanity-check a human valuation against the model. The consumer-facing price estimates on property portals are AVMs too. The term is used on both sides of the Atlantic; in the UK AVMs are common on low loan-to-value remortgages, and in the US their use in lending is regulated, with quality control standards for AVMs used in credit decisions adopted by the federal agencies in 2024.
How an automated valuation model works
- Data in. Recorded sales, tax and title records, property attributes, listings, and price indices. The model is only as current and complete as these feeds.
- Modelling. Most AVMs blend approaches: comparable-sales logic that selects and adjusts recent sales, hedonic regression that prices attributes such as floor area and plot size, and index models that roll a previous sale price forward with market movement.
- Value out, with a confidence score. A good AVM states how sure it is, often as a forecast standard deviation or a confidence grade. In a uniform estate with many recent sales the band is tight; for an unusual property in a thin market it is wide.
- Use rules decide what happens next. Lenders set thresholds: accept the AVM below a given loan-to-value and above a given confidence, otherwise escalate to a desktop valuation, a hybrid appraisal or a full physical valuation.
How AVMs are used
- Lending decisions on low-risk cases: remortgages, further advances and home equity lending where equity is comfortable and the property type is common.
- Portfolio monitoring: revaluing a loan book monthly or quarterly for risk and capital reporting.
- Quality control: flagging human valuations that sit far from the model's estimate for a second look.
- Triage: deciding which cases deserve the cost of a visit, so the physical valuations go where they change the decision.
AVM example: a home equity application
A homeowner applies for a home equity line against a standard townhouse, purely as an illustration. The lender's AVM returns $410,000 with a high confidence grade: the model found eleven sales of near-identical houses in the same development within a year, and the previous sale of this house two years ago fits the pattern. The requested credit keeps total borrowing at 60 percent of that value, inside the lender's AVM acceptance rules, and the application proceeds the same day without a valuation appointment.
The next applicant on the same street is declined the AVM route: her house has a large extension the records do not show, the model's confidence grade is low, and the case escalates to a valuation where a person looks at the property. The model did its job both times, once by answering and once by knowing it could not.
What an AVM does not do, and the mistakes people make
It cannot see condition. The model prices the property the records describe: the renovated house and the fire-damaged house next door, identical on paper, get the same number. This is the structural blind spot of every AVM.
It is not an appraisal or a valuation in the professional sense. No qualified person has formed or signed an opinion. A desktop valuation has a valuer behind it; an AVM does not, and lenders and regulators treat the two differently.
It does not inspect, measure or verify anything. Errors in the source records pass straight through into the estimate.
The common mistakes: using an AVM on unusual properties or thin markets where the confidence band is wider than the decision; ignoring the confidence score and treating every estimate as equally solid; letting stale records price a property that has been extended, subdivided or damaged; and consumers treating a portal estimate as a promise of what a buyer will pay. The number is a model output, not a market outcome.
Where the output goes
Into the credit file as the valuation evidence for that decision, with the model name, version, date and confidence score recorded, because regulators and auditors ask which model said what and when. In portfolio use the outputs feed risk reporting rather than individual files. Where the AVM's confidence is too low, the case moves up the ladder: desktop valuation, hybrid appraisal or a full visit, each adding human judgement the model does not have.
The one input an AVM never has is current condition, and that gap can be filled without a visit. Venta Capture, a product of VentaVid, is one way to do it: the occupier receives a link and follows guided steps in the mobile browser with no app, room by room, and the photos and video arrive timestamped, located and digitally sealed on a case. Venta Capture is not a valuation model and estimates nothing; it supplies the current condition evidence an AVM lacks, for the lender or valuer who makes the decision to review alongside the model's number. See Venta Capture.
Try Venta Capture on your own process
Build one flow for your highest-volume case type. Free, no credit card.