An investor screening forty listings before breakfast cannot commission forty appraisals. They need a number for each in seconds, accurate enough to decide which properties are worth a closer look. That number comes from an automated valuation model.
AVMs have quietly become part of how property gets bought, lent against and analysed. They sit behind the estimate on a listing portal, inside a lender's underwriting flow, and across an investor's screening pipeline.
This article explains what an AVM is, how the AI behind it turns data into a value, how accurate those values are, and where a human still has to step in.
What Is an Automated Valuation Model? (AVM Explained)
An automated valuation model is a software system that estimates a property's market value using data and statistical modelling, without a person visiting the home. It takes an address, pulls the data it holds on the property and its surroundings, then returns an estimated value in seconds.
The estimate is a probability, not a fact. It answers what a property would likely sell for in current market conditions, based on the patterns the model has learned from past sales. Its quality depends entirely on the data and the model behind it.
Most AVMs return more than a single number. Alongside the estimate, a well-built model gives a value range and a confidence score signalling how reliable it is. A high score means strong data; a low one is a flag to treat the number with caution.
How AI Processes Data to Generate Property Values
An AVM turns raw data into an estimate through a sequence of steps that varies in detail by provider but keeps a consistent shape.
1. Cleaning the data. Raw property and transaction records contain errors and duplicates. They also hold sales that do not reflect true market value, such as foreclosures, family transfers and other non-arm's-length deals. ATTOM describes validating and removing these before any modelling begins, because a valuation built on bad inputs is worse than no valuation at all.
2. Adjusting old sales to present conditions. A sale from three years ago cannot be used at its original price, because the market has moved since. Modern AVMs build hyperlocal price indices, tracking how prices have shifted within a small area, and use them to translate historical sales into today's money.
3. Generating the valuation. This is where the AI works. Rather than one formula, a strong model runs several machine-learning models together, each capturing a different aspect of value. It blends them based on the data available for that property, producing a single estimate with its confidence score attached.
Two modelling approaches sit underneath most systems. Hedonic models price a property by its features, valuing each bedroom, square foot and location attribute. Repeat-sales models track how the same properties change hands over time to read the price trend. Many combine both.
The Data Inputs: Sales Comps, Tax Records, Location, and Market Trends
An AVM is only as good as the data feeding it. Four categories do most of the work.
- Comparable sales: recent sales of similar nearby properties, the single most important input. Comps tell the model what buyers have actually paid for homes like this one.
- Property characteristics: the physical facts of the home, drawn from public records. Bedrooms, bathrooms, square footage, lot size, year built and property type all shape the estimate.
- Tax and assessor records: county assessor and recorder data, giving the model an official baseline for ownership, past sale prices and assessed value.
- Location and market trends: where the property sits and where the local market is heading. Neighbourhood, school districts, and current supply and price direction all move the number.
The best models pull these from a broad, validated national dataset covering the whole country. Breadth matters, because the more comparable data a model can draw on, the more reliable its estimate for any single home.
How Accurate Are AI Valuations? (Margin of Error Analysis)
Accuracy decides whether an AVM is useful, and the honest answer is that it varies by provider, property and market.
Two published figures give a sense of the range. Zillow reports a national median error rate for its Zestimate on off-market homes, its published measure of how far the estimate typically lands from sale price. ATTOM reports its AVM hitting a median absolute percentage error of 2.9%, with about 80% of valuations within 10% of the sale price and 92% within 20%.
Read those numbers carefully. A median error means half of estimates are closer and half are further, so any single valuation can be well outside the headline figure.
Accuracy also depends on conditions. AVMs are strongest where data is plentiful, in dense suburban markets full of similar homes that trade often. They are weakest where data is thin, in places like unique properties, rural areas, custom builds and low-transaction markets with few recent sales.
An AVM estimate is a fast starting point rather than a settled figure, and the confidence score tells you how much weight it can bear.
AVM vs Formal Appraisal: When to Use Each
An AVM and a formal appraisal answer the same question by opposite methods, each fitting a different job.
An AVM is a software estimate produced in seconds from data, with no property visit. It is fast, cheap and consistent, and scales to thousands of properties at once. It cannot see the inside of the home, the quality of a renovation, or the damp patch in the basement.
A formal appraisal is a valuation by a licensed appraiser who inspects the property in person. It is slow and expensive by comparison, taking days and costing hundreds of dollars. In return it captures the condition, quality and unique features no dataset holds. For a mortgage on a specific home, it carries a legal weight an AVM does not.
The choice follows the stakes and the volume. Use an AVM when speed and scale matter and the cost of being slightly wrong on any one property is low, such as screening a portfolio or setting an asking-price range. Use an appraisal when a single decision carries real financial or legal weight, such as closing a mortgage. Many workflows use both, an AVM to triage and an appraisal to confirm.
How Property Investors Are Using AVM Tools in 2026
For investors, the value of an AVM is speed at scale. It turns a task that used to take days into one that takes minutes, and it changes how a deal pipeline runs.
The most common use is screening. An investor evaluating dozens of properties runs each through an AVM to sort them fast, spotting the ones priced below their estimated value and discarding the rest. What once needed a week becomes a first pass in an afternoon.
Portfolio monitoring is the second use. An owner holding many properties can run the whole set through an AVM periodically to track how values are moving, without commissioning a valuation on each. Rental AVMs do the same for income.
The discipline that separates good use from bad is knowing the limits. An investor who treats an AVM as a screening tool, then does real diligence on the shortlist, gets the speed without the risk. One who skips the closer look is trusting a number the model itself only offered as a probability.
The Best AVM Tools Available for US Real Estate
There is no single best AVM; the right one depends on who is asking and why. The market splits roughly into three groups, and the question is which fits your use case rather than which ranks highest.
- Consumer portals: Zillow's Zestimate and similar consumer-facing estimates are free, instant and useful for a rough sense of value. They are built for a broad audience, so they suit casual checks more than professional decisions.
- Data and analytics providers: ATTOM, Cotality (formerly CoreLogic), Clear Capital and First American offer lender-grade and investor-grade AVMs with wider coverage, confidence scoring and API access. These are built for professional workflows and priced accordingly.
- Investor-focused tools: platforms aimed at investors bundle AVM estimates, including rental AVMs, into screening and portfolio tools, so the valuation sits alongside the rest of a deal workflow.
When comparing options, look past the headline value. Check the coverage in your target markets and the transparency of the confidence score. Check the accuracy figures the provider publishes and whether the delivery method fits how you work. A model accurate on paper but thin in your market, or opaque about its confidence, will not serve you well.
Where Valuation Ends and VerbaFlo Begins
An AVM answers one question, telling you what a property is worth when you decide to acquire or hold it. That number matters, but it is a starting figure. Whether the value holds depends on how the property performs once people are living in it, and that is a different discipline entirely.
Occupancy and retention turn an estimated value into a realised one, and those outcomes rest on communication. A vacant unit, or a resident who leaves over a slow response, erodes the value an AVM measured on paper.
That operational layer is where VerbaFlo works. It is a conversational AI platform for residential real estate, handling enquiries, qualification and bookings across voice, chat, WhatsApp and email. The AVM sizes the opportunity; running the property well protects it.
