Automated Valuation Models (AVMs) Explained

AVMs blend several techniques. Hedonic regression gives a price to each feature, such as square footage, bedrooms, lot size and location. Repeat-sales indexes track how prices of the same homes change between sales, then apply that growth to a property's last sale price. Comparable-sales engines copy what appraisers do by finding similar nearby sales. Modern AVMs often feed hundreds of variables into machine learning models such as gradient-boosted trees or neural networks, and many combine several models into an ensemble. Zillow launched the Zestimate in 2006 and publishes its own median error figures. Error is much lower for homes currently listed for sale than for off-market homes, because the list price, fresh photos and description give the model strong new information. A median error also means half of all homes are off by more than that figure. Accuracy varies by market. Tracts of similar homes with frequent sales are easiest to value. Rural properties, unusual or luxury homes, fast-moving markets and non-disclosure states such as Texas, where sale prices are not recorded publicly, are harder. A common misconception is that an AVM equals an appraisal. An AVM usually cannot see a renovated kitchen, a failing roof or a noisy road unless those facts show up in the data. Zillow's exit from home-buying in 2021 is often cited as a warning about how hard it is to price homes algorithmically when markets shift. Regulation has caught up. In 2024, six federal agencies finalized an AVM quality control rule, effective October 1, 2025, for mortgage originators and secondary market issuers that use AVMs in credit decisions on a consumer's principal dwelling. It requires policies designed to ensure a high level of confidence in estimates, protect against data manipulation, avoid conflicts of interest, require random sample testing and reviews, and comply with nondiscrimination laws. Consumer browsing tools like the Zestimate are outside its scope.