An automated valuation model (AVM) is software that estimates a property’s market value in seconds by running public records, recent sales and local market data through statistical and machine learning models. So if you are wondering what is AVM in real estate, the short answer is that it is the engine behind the instant home values you see on Zillow, Redfin and your lender’s portal, and more and more often it sits behind real lending decisions in the US too.
That second part is why AVMs deserve more attention than they usually get. Since October 1, 2025, mortgage originators and secondary market issuers that use AVMs in covered credit decisions have had to follow federal quality control standards. Lenders now approve some home equity lines and refinances with no appraiser on site at all, and a growing number of proptech and fintech startups are building valuation features straight into their products.
This guide walks through what an automated valuation model is, what goes into one, how AI has changed the way property gets priced, how accurate the numbers really are, and how AVMs compare with RVMs, CMAs and full appraisals. If you are a founder or technical lead thinking about building valuation into your own platform, the later sections cover the build steps and the options for getting it done.
What Does AVM Stand For in Real Estate?
AVM stands for automated valuation model. It is a computer model that produces an estimated market value for a specific property as of a specific date, without anyone visiting the home. The estimate usually arrives with a value range and a confidence score so the person using it knows how much to trust the number.
Most AVMs in the US are built on two classic techniques, often blended together.
- Hedonic models treat a home as a bundle of features. Square footage, bedrooms, bathrooms, lot size, age and location each get a price based on regression analysis of recent sales, and the model adds them up for the subject property.
- Repeat sales indexes look at homes that have sold more than once, measure how prices in that area moved between sales, and apply that trend to the subject property’s last recorded price.
A third family, comparables based AVMs, picks a set of similar nearby sales for each property and adjusts them for differences, much like an appraiser would. Because the comps are chosen per property, the result can be traced back and explained, which lenders appreciate.
Commercial grade AVMs rarely rely on a single method. They run several models, weight each result by how confident it is, and report one blended estimate. Freddie Mac’s Home Value Explorer, for example, has long combined a repeat sales model with a hedonic model to arrive at its final value.
What Is Usually Included in an AVM Valuation?
A typical AVM report includes an estimated value, a high and low range, a confidence score, the effective date and the comparable sales the model leaned on. The inputs behind that report usually come from five places.
- Property characteristics: living area, lot size, bedrooms, bathrooms, year built, property type and sometimes features like a pool or garage.
- Transaction history: prior sale prices and dates for the subject home and its neighbors, pulled from county deed records.
- Tax assessments: assessed values and the property details recorded by the local assessor.
- Listing data: active, pending and recently sold listings from the MLS where the provider has a license to use them, including list prices and days on market.
- Market trends: price indexes at the ZIP code, neighborhood or metro level that show which way values are moving.
What an AVM does not include matters just as much. It has no idea that the kitchen was redone last year, that the roof leaks or that the house backs onto a busy road unless that information shows up somewhere in the data. That gap is the main reason AVM estimates can drift away from what a buyer would actually pay.
How AI Is Pricing Property Today
Ten years ago, the honest answer to what is AVM in real estate would have been a regression formula sitting on top of county records. That is no longer the full picture. The best AVMs on the market now use machine learning at almost every step, and the gains come mostly from reading data that older models simply ignored.
- Machine learning models in place of single formulas. Gradient boosted trees and neural networks can pick up patterns a linear model misses, such as a corner lot being worth more in one suburb and less in the next. Zillow moved its Zestimate to a neural network model in 2021 for exactly this reason.
- Computer vision on listing photos. Models can now score finish quality, natural light and visible condition from photos, which narrows the old blind spot around renovations. Brookings has noted that AVM developers are adding computer vision to improve their estimates.
- Language models on listing text. Phrases like “new roof”, “original kitchen” or “sold as is” carry real price signals, and natural language processing turns them into model features.
- Faster refresh cycles. Instead of monthly updates, many providers now refresh values weekly or even daily as new sales and listings come in.
AI does not remove every weakness, though. A home that has not been listed in years has no fresh photos or description to read, so off market estimates still lean on older data. Machine learning can also absorb historical bias from the sales it learns from, which is part of why regulators now expect lenders to test AVMs for fair lending problems. Teams adding valuation to a product should plan for explainability from day one, and the same thinking applies if you are wiring AI agents into your workflows around those valuations.
When Did Automated Valuation Models Emerge?
Automated valuation models grew out of academic pricing research in the 1970s and 1980s and moved into US mortgage lending through the 1990s. The table below traces the key moments, from the theory behind them to the federal rule that now governs their use.
If you are looking for seminal papers on automated valuation models, Rosen (1974) and Case and Shiller (1987) are the two most cited starting points.
AVM vs RVM vs CMA vs Appraisal: What Is the Difference?
An AVM is a computer estimate, an RVM is a Realtor-only AVM that adds MLS data, a CMA is an agent’s pricing analysis and an appraisal is a licensed professional’s opinion of value. Anyone who searches what is AVM in real estate tends to run into all four terms within a few minutes, so it helps to see them side by side.
What is RVM in real estate? RVM stands for Realtors Valuation Model. It is the automated valuation product offered through Realtors Property Resource (RPR) and owned by the National Association of Realtors, and only Realtors can access it. Where a standard AVM in RPR draws only on publicly recorded sales, the RVM also feeds in on market and off market MLS listing data, which is why RPR describes it as its most accurate estimate. RPR refreshes both RVM and AVM values twice a month.
What is a CMA? A comparative market analysis is a report a real estate agent prepares by hand, usually to set a list price. The agent picks the comps, walks or knows the property, and adjusts for condition in a way no model can fully match.
What is an appraisal? An appraisal is a formal opinion of value from a licensed or certified appraiser who inspects the home. It is still the standard for most US purchase mortgages.
How Accurate Is an AVM in Real Estate?
AVMs are quite accurate for typical homes that are on the market and much less accurate for off market or unusual properties. Zillow’s own published figures put the Zestimate’s nationwide median error at roughly 2% for homes listed for sale and around 7% for homes that are not, and the off market figure can climb past 12% in states with thin sales data. On a $500,000 house, a 7% miss is about $35,000 in either direction.
The gap between listed and unlisted homes is easy to explain. Once a home is listed, the model can see the asking price, the photos and the description, and it adjusts toward them. An off market home only has whatever the county and the last sale recorded.
What an AVM confidence score tells you
Most providers attach a confidence score to each estimate, and it is worth reading before you rely on the number. RPR’s scale is a useful example because it publishes exactly what each level means.
A high score means several models agree on the value. A low score usually points to missing property facts, few recent sales nearby or a home that does not look like its neighbors. Rural properties, luxury homes, newly built homes and anything heavily renovated tend to land at the lower end.
How AVMs Are Used in Mortgage Lending in the US
Lenders use AVMs to approve home equity products, grant appraisal waivers on some refinances and purchases, check appraisals for red flags and watch the value of the loans they already hold. Home equity is where most borrowers meet an AVM directly. A HELOC valued by AVM often costs the borrower nothing or less than $75, compared with roughly $350 to $800 for a full appraisal, and some lenders only accept AVM valuations for home equity loans.
On the agency side, Freddie Mac’s Automated Collateral Evaluation decides whether an AVM estimate can stand in for an appraisal on eligible loans, and Fannie Mae runs a similar waiver program. When Fannie Mae or Freddie Mac uses an AVM to offer that waiver, the compliance duty under the new federal rule sits with them rather than the originating lender.
The federal AVM quality control rule
The final rule on quality control standards for automated valuation models, issued by the CFPB together with the OCC, Federal Reserve, FDIC, NCUA and FHFA, took effect on October 1, 2025. It applies when mortgage originators and secondary market issuers use an AVM to value a consumer’s principal dwelling for a credit decision or a covered securitization decision. Using AVMs only to monitor existing loans or review completed appraisals falls outside it.
Covered institutions must keep policies, procedures and controls that do five things:
- Ensure a high level of confidence in the estimates the AVM produces.
- Protect against manipulation of the data the AVM uses.
- Avoid conflicts of interest.
- Require random sample testing and reviews.
- Comply with applicable nondiscrimination laws.
For anyone building lending or proptech software, the practical effect is that a valuation model now needs logging, testing and documentation an examiner can follow, not just a good accuracy number.
How to Build a Custom AVM for Your Proptech or Lending Product
Once your team understands what is AVM in real estate, the next decision is whether to license a model from a vendor or build a custom AVM that fits your own data and users. Licensing is quicker to switch on, but a custom build lets you control accuracy in your target markets, own the model and explain every estimate to a regulator or a customer. The steps below describe how a custom AVM build usually runs for a US startup.
- Define the use case and accuracy target. A consumer home value widget, a HELOC decision engine and an investor underwriting tool need very different error tolerances. Write down the markets, property types and median error you need before anyone writes code.
- Secure your data sources. Line up public record and deed data, tax assessor data and, where you qualify, MLS listing feeds. Data licensing often takes longer than modeling, so start it first.
- Clean, match and geocode records. Merge sales, listings and property facts to a single property ID, fix missing square footage and flag outlier sales like family transfers or foreclosures.
- Build baseline models. Start with a hedonic model and a repeat sales index so you have an explainable benchmark to beat.
- Add a machine learning layer. Train gradient boosted or neural models on the cleaned data and test image and listing text features where you have them.
- Produce ranges and confidence scores. Blend the models, then calibrate a value range and score for each estimate so users know when to trust it.
- Validate, test for bias and document. Hold out recent sales by market, measure median error, check results across neighborhoods for disparate outcomes and keep the records the federal rule expects.
- Ship it as an API and keep it fresh. Deploy behind a versioned API, monitor drift as new sales arrive and retrain on a fixed schedule.
This is a data engineering and machine learning project as much as a real estate one, so it pays to bring in people who have built custom AI software for production before. Many founders start with a focused proof of concept in one metro through a custom product development team, then expand market by market once the error numbers hold up.
Best options for building an AVM in the USA
If you would rather license than build, the top automated valuation model companies serving US lenders include CoreLogic, ICE Mortgage Technology, Clear Capital, HouseCanary and Veros. If you want your own model, here is how the main build options compare.
Rocketeams sits at the top of that list for most startups because it combines a custom build with the lowest cost. Teams are made up of top 1% engineers from South Asia, AI matched to the project and fully managed from sprint planning through payroll, at roughly a third of US rates. If you only need a couple of extra people on an existing data team, you can also hire dedicated software developers individually. Valuation work also overlaps heavily with lending, payments and compliance, which we cover in our guide to choosing a fintech software development company in the USA.
Frequently Asked Questions
How accurate is AVM in real estate?
For homes that are listed for sale, leading AVMs like the Zestimate have a nationwide median error of about 2%. For off market homes the median error rises to around 7%, and it can be much higher for rural, luxury or recently renovated properties.
How much does an AVM appraisal cost?
Consumer AVMs on sites like Zillow and Redfin are free. Lenders typically pay under $100 per report, and borrowers often pay nothing or less than $75 when a HELOC is valued by AVM, compared with about $300 to $600 for a traditional appraisal on a single family home.
Is an AVM an appraisal?
No. An AVM is a computer generated estimate with no site visit, while an appraisal is a licensed professional’s opinion of value based on an inspection. People sometimes say “AVM appraisal”, but most purchase mortgages still require a real appraisal unless the loan qualifies for an appraisal waiver.
How long does an AVM appraisal take?
An AVM returns a value in seconds. A traditional appraisal usually takes anywhere from a few days to a few weeks from order to final report.
How do you get an AVM appraisal?
Homeowners can get a free estimate from consumer sites by entering an address. For lending, the lender orders the AVM through a vendor or its own system, and borrowers usually see the result only if the lender shares it.
What are the key differences between an appraisal and an AVM?
The main differences are inspection, cost and speed. An appraisal includes a physical inspection, costs a few hundred dollars and takes days, while an AVM sees no interior, costs little or nothing and is instant.
What are the three valuation methods for real estate?
The three standard approaches are the sales comparison approach, the cost approach and the income approach. Most residential AVMs automate the sales comparison approach, while income based models are more common for rentals and commercial property.
What is RVM in real estate?
RVM stands for Realtors Valuation Model, an AVM available only to Realtors through Realtors Property Resource. It adds on market and off market MLS data to public sales records, which usually makes it more accurate than a standard AVM.
What is the difference between an AVM and a CMA?
An AVM is produced entirely by software, while a CMA is prepared by a real estate agent who picks comps and adjusts for condition. A CMA is usually the better guide for setting a list price.
Is there a commercial AVM?
Yes, commercial AVMs exist for property types like multifamily, retail and office. They are harder to build because commercial sales are fewer and value depends heavily on income and leases, so their estimates usually carry wider ranges.
What is AVM in real estate used for besides lending?
Beyond mortgage decisions, AVMs are used for property tax assessment support, insurance underwriting, investor portfolio valuation, iBuyer offers and the home value features inside real estate apps.
Conclusion
Automated valuation models have moved from a back office tool for lenders to something buyers, sellers, agents and regulators all deal with. They are fast and cheap, they are getting sharper as AI reads photos and listing text, and they still struggle with homes that are off the market, unusual or recently renovated. The federal quality control rule now in force means any AVM used in a mortgage decision needs real testing and documentation behind it.
For US startups, that creates room to build better valuation products for specific markets and users. If you are planning a custom AVM or a valuation feature inside your app, Rocketeams can put a fully managed team of top 1% engineers on it in under 100 hours, at about a third of US rates, with a two week risk-free trial. And the next time someone asks you what is AVM in real estate, you can tell them it is software that prices property from data, and that the best ones are now powered by AI.
