US Real Estate & Mortgage Research Analyst
Published: August 14, 2026 · Updated: August 14, 2026 · 11 min read
Key Takeaways
- • An AVM is a statistical algorithm, not a physical inspection — Zillow and Redfin publish their accuracy, but the models estimate, they do not appraise
- • Published median error: Zestimate ~1.9% on-market / ~7.0% off-market; Redfin Estimate 2.01% on-market / 7.73% off-market
- • "Median error" means half of estimates miss by more than that — so a typical off-market miss is about $35,000 on a $500,000 home
- • Six federal agencies adopted AVM quality-control standards (Federal Register Aug 7, 2024; effective Oct 1, 2025) covering mortgage-use valuations
- • AVMs miss condition, lot appeal, views, and fast market shifts — use them as a starting range, then confirm with comparable sales, a CMA, or a licensed appraisal
Type any address into Zillow or Redfin and within seconds you get a dollar figure for what the home is "worth." These instant numbers — the Zestimate and the Redfin Estimate — are the most visible examples of automated valuation models (AVMs): algorithms that estimate home value from data rather than a person walking through the property. They are genuinely useful, genuinely popular, and genuinely imperfect. This guide explains how they work, the accuracy the companies actually publish, the federal rules now governing them, and exactly where to use — and not use — an online estimate.
What is an automated valuation model?
An AVM is software that produces an opinion of value by crunching large datasets. According to the companies, the typical inputs include:
- Public records: Tax assessments, prior deed and sale prices, and county permit data.
- Property characteristics: Square footage, lot size, year built, bedroom and bathroom counts, and home type.
- Comparable sales: Recent sales of similar nearby homes — the same "comps" concept an appraiser uses, but applied statistically.
- Listing data (on-market): For active listings, the list price, photos, descriptions, and days on market sharpen the estimate.
- User-submitted facts: Homeowners can correct bed/bath counts or report renovations, which may improve accuracy.
Zillow states it has issued Zestimates for more than 104 million homes nationwide (Investopedia, 2025). The model is refined continuously and, for on-market homes, updated frequently. The crucial limitation is the same for every AVM: it values the property from data it has, never from a visit. It assumes "average" condition and cannot see a renovated kitchen, a failed roof, or a busy-street location the way a human can.
The published accuracy numbers
Both major platforms publish a median error rate — the share by which the estimate typically misses the actual sale price, with half of estimates closer and half farther. The figures below are the companies' own published numbers:
| Platform | On-market median error | Off-market median error | Source |
|---|---|---|---|
| Zillow Zestimate | ~1.9% (1.94% per Zillow data) | ~7.0% (7.06%) | zillow.com/zestimate via Investopedia |
| Redfin Estimate | 2.01% | 7.73% | redfin.com/redfin-estimate |
Two things jump out. First, for homes actively listed for sale, the models are reasonably tight — within about 2% of the sale price half the time. Second, for off-market homes (which describes the vast majority of owner estimates, since most people are just "checking"), the median error roughly triples. As Redfin notes, on-market estimates are more accurate precisely because there is more and fresher data about those homes (Redfin).
What that error means in real dollars
A percentage is abstract; the dollar swing is not. On a $500,000 home, a 7% median off-market error implies a typical miss of about $35,000 — and remember, "median" means half of all estimates are off by more than that. On a $1,000,000 home, the same 7% is about $70,000. For a buyer making an offer or a seller setting a list price, that gap is the difference between a sound decision and a costly one.
Federal oversight: the 2024 AVM rule
AVMs are not a regulatory free-for-all when they touch a mortgage. On August 7, 2024, six federal agencies — the Office of the Comptroller of the Currency, the Federal Reserve Board, the FDIC, the National Credit Union Administration, the Consumer Financial Protection Bureau, and the Federal Housing Finance Agency (FHFA) — adopted a final rule establishing quality-control standards for AVMs used to value homes securing a consumer's principal dwelling (FHFA / Federal Register, 89 FR 64538). The rule became effective October 1, 2025.
Under the rule, institutions using AVMs in covered credit or securitization decisions must adopt policies and controls that:
- Ensure a high level of confidence in the estimates produced
- Protect against manipulation of data
- Seek to avoid conflicts of interest
- Require random-sample testing and independent reviews
- Comply with applicable nondiscrimination laws
Importantly, this rule governs AVMs used inside the mortgage system (lenders, Fannie Mae, Freddie Mac, and securitization). The consumer-facing Zestimate and Redfin Estimate you see for free are not themselves "appraisals," but the rule reflects how seriously regulators now treat model-based valuation. It also underscores a theme of this guide: an estimate is only as trustworthy as the data, controls, and validation behind it.
What an AVM can't see
No algorithm visits the property, so several value drivers are systematically missed or assumed average:
- Condition and finish quality: A custom kitchen and a dated one may carry the same bedroom/bath count to a model.
- Deferred maintenance: An aging roof or failing HVAC is not in public records until it becomes a permit or sale.
- Lot and location nuance: A quiet cul-de-sac vs. a busy road, a view vs. a wall, a corner lot vs. interior — these move value but are hard for models to weight.
- Renovations not in public data: Finished basements, added baths, or landscaping only help if someone reports them.
- Market timing: AVMs lean on historical sales and can lag quickly shifting markets, over- or under-shooting current demand.
This is exactly why both companies frame their estimates as a "starting point, not an appraisal." A real estate agent's Comparative Market Analysis (CMA) and a licensed appraisal exist to capture the human, on-site factors an AVM cannot.
How AVMs fit with appraisals and CMAs
Think of valuation as a ladder of precision and cost:
- AVM (free, instant): A ballpark range. Great for "is this closer to $300k or $350k?" and for tracking a band over time.
- CMA (free, 1–3 days): An agent's comparable-sales analysis that adds local knowledge and often a walkthrough.
- Licensed appraisal (paid, ~1–2 weeks): The documented, USPAP-compliant opinion lenders require for financing — the gold standard for transactions.
For market context at the metro or state level, the FHFA House Price Index remains the authoritative public source for how prices have moved, complementing — not replacing — a single-home estimate. Our Home Value Appreciation Calculator is a planning tool for modeling a range of scenarios; it is never a forecast of your specific home's value.
A practical rule of thumb
Use an online estimate to set expectations and sanity-check a range. Then, before any money moves, validate that range against three to five recent comparable sales, get a free CMA from a local agent, and — for a purchase, sale, refinance, or loan — rely on a licensed appraisal. Cross-referencing two AVMs (Zillow and Redfin) plus comparable sales is far more reliable than trusting either number alone.
Authoritative sources used in this guide
- Zillow — About Zestimate (published median error rates, on-market and off-market): zillow.com/zestimate
- Redfin — About the Redfin Estimate (median error 2.01% on-market, 7.73% off-market): redfin.com/redfin-estimate
- Investopedia — comparison of Redfin and Zillow estimate accuracy and Zillow's home coverage: investopedia.com
- Federal Housing Finance Agency (FHFA) / Federal Register — Quality Control Standards for Automated Valuation Models, 89 FR 64538, effective Oct 1, 2025: fhfa.gov
- Federal Housing Finance Agency (FHFA) House Price Index — public market-wide price trends: fhfa.gov/DataTools/Downloads/Pages/House-Price-Index.aspx
Sources retrieved: August 14, 2026.
Editor Update Note
This article was reviewed and published on August 14, 2026. Source links (Zillow Zestimate, Redfin Estimate, Investopedia, FHFA AVM final rule, FHFA House Price Index) were verified as live at the time of publication. Accuracy figures reflect the companies' published median error rates as of mid-2025; they are subject to change as models are updated. Methodology descriptions follow standard U.S. appraisal and AVM practice and are explanatory only — they are not a valuation of any property.
Try These Free Calculators
- Home Value Appreciation Calculator — model a range of long-term value scenarios for planning.
- Mortgage Payment Calculator — estimate PITI and compare fixed vs. ARM terms.
Frequently Asked Questions
What is an automated valuation model (AVM)?
An AVM is a computer algorithm that estimates a home's market value by analyzing public records, tax assessments, prior sales, and recent comparable sales, often blended with listing data for homes that are on the market. Zillow's Zestimate and Redfin's Redfin Estimate are the two AVMs that publish their accuracy figures for U.S. consumers. An AVM is a statistical estimate, not a physical inspection, and it is not an appraisal.
How accurate are Zillow Zestimate and Redfin Estimate?
According to figures the companies publish, the Zestimate has a median error rate of about 1.9% for homes that are currently listed for sale (on-market) and about 7.0% for off-market homes. The Redfin Estimate reports a median error rate of 2.01% for on-market homes and 7.73% for off-market homes. "Median error" means half of all estimates are within that percentage of the eventual sale price and half are farther off — so the typical off-market miss is roughly $35,000 on a $500,000 home.
Why are off-market estimates so much less accurate?
When a home is listed for sale, the AVM can use fresh, verified listing data — list price, photos, descriptions, and days on market. For an off-market home, the model falls back on older public records such as tax assessments and the last recorded sale, which may not reflect renovations, deferred maintenance, or current market conditions. That thinner, staler data is why the median error roughly triples for off-market properties.
Are online estimates regulated?
Yes, when they are used in mortgage decisions. Six federal agencies — the OCC, Federal Reserve, FDIC, NCUA, CFPB, and FHFA — adopted a final rule setting quality-control standards for AVMs used in valuing homes that secure a consumer's principal dwelling. Published in the Federal Register on August 7, 2024 (89 FR 64538), the rule became effective October 1, 2025. It requires institutions to ensure a high level of confidence in estimates, guard against data manipulation, avoid conflicts of interest, test models, and comply with nondiscrimination laws.
Can an AVM replace a professional appraisal or CMA?
No. AVMs are useful for a quick ballpark and for tracking a range, but they cannot see interior condition, finished basements, lot appeal, views, or neighborhood nuances, and they lag fast-moving markets. For any purchase, sale, refinance, or borrowing decision, lenders require a licensed appraisal and agents rely on a Comparative Market Analysis (CMA). Use an AVM as one input, then validate it against comparable sales and a professional opinion.