AI Case Study

Zillow provides real estate price estimates based on new transaction data through machine learning

Real estate marketplace Zillow uses machine learning and big data to provide a more specific estimate of a home's price, called the Zestimate. This is a marketing tool to drive users to the Zillow site and the pricing has a median error rate of 4.6%.


Financial Services

Real Estate Development Operations

Project Overview

According to Zillow: "The Zestimate® home valuation is Zillow's estimated market value, computed using a proprietary formula. It is not an appraisal. It is a starting point in determining a home's value. The Zestimate is calculated from public and user-submitted data, taking into account special features, location, and market conditions... a team of statisticians is working every day to make the Zestimate more accurate. Since Zillow's inception in 2006, we have deployed three completely new versions of the algorithm (2006, 2008 and 2011), but incremental improvements are made between major upgrades with new iterations being deployed regularly."

Reported Results

Zillow claims its Zestimate "accuracy has a median error rate of 4.6%. This means half of the home values in the area are closer than the error percentage."


Details undisclosed



Sales Operations


From Zillow: "The Zestimate home valuation was also designed to be independent of any opinion from either the seller or buyer. Neither party is a neutral, unbiased observer in the transactions. We’d like to arrive at an estimate of value that is neutral to the opinions on either side of the transaction." Previously the approach to valuing homes was done by statisticians for geography and time periods (more art than science). Now machine learning models developed constantly to evaluate homes nightly based on daily transactions and new listings, making use of a large amount of data.



Public and user-submitted data. From Zillow: "Zestimate modeling framework, which contains numerous submodels, each estimating a home’s value via different valuation approaches and data inputs, our goal of independence means that the listing price is not a factor in any of our valuation submodels".