Skip to content


Zillow made a few announcements today including the fact that it has reached the start-up rite of passage (or sometimes-PR stunt) of coming out of beta. But more important than this symbolic move are the reasons behind it.

The company announced that it has grown its database of homes to about 80 million in 48 states. This is 27 million more than it had when it launched two years ago and represents 88 percent coverage of U.S. homes. By coverage, I mean it has applied its flagship Zestimate home value to these properties.

Zillow also announced that it has improved its Zestimate accuracy by incorporating a 20x increase in the number of models that are used to derive home value estimates. These include user-submitted home facts (number of bedrooms, bathrooms, additions, renovations, etc.) as well as algorithms that are more geographically specific.

“We’ve had about a million users edit their home facts totaling about 16 million attributes. For the first time we’re plugging those 16 million new data points into our models,” says Zillow CFO Spencer Rascoff. “The other driver of accuracy is more models. Instead of models for each county, we’ve taken it down to the neighborhood level. That’s 334,000 unique models around the country.”

These changes have already resulted in a 12 percent accuracy increase nationwide and higher increases in top real estate markets such as San Francisco (18 percent), Miami (21 percent), Los Angeles (22 percent) and Atlanta (28 percent). Accuracy is measured by the difference in home sales and the Zestimate value the day before the sale (more accuracy stats here).

This accuracy, and its effective communication, is important for Zillow given the misgivings some people have toward an automated home estimate tool. Still the site has had 40 percent of U.S. homes searched since it launched in 2006, and it receives more than 25,000 community or home-owner contributions per day.

Accuracy is also important if you consider that Zestimates are essentially a cornerstone of the company’s product model.

“Zestimates are how we got here,” says Rascoff. “They are what we are most associated with from a brand perspective. So increasing our accuracy and data coverage goes straight to the traffic we hope the Web site will get. It can always stand to be improved.”

This Post Has One Comment

Leave a Reply

Back To Top