Elliot Njus The Oregonian
Zillow’s computer algorithm has developed an eye for design.
The real estate website, built around providing a value estimate for just about every home, says its algorithm got an upgrade that lets it detect granite countertops, hardwood floors or large windows that let in natural light. The photos come from previous sale listings as well as user-uploaded photos.
And it’s using that information to inform its pricing estimates, which are also based on property records, listing data and sales of other similar houses.
The technology isn’t yet sophisticated enough to actually identify the features. Instead, it identifies certain pixel patters and assigns a corresponding change in value. The company trained its algorithm using the database of millions of home photos already uploaded to the site.
With the upgrade, Zillow says half of its estimates fall within 2% of the price for which a home eventually sells.
Some homeowners and real estate brokers have long been frustrated with Zillow’s automated appraisals, which they say can set unreasonable expectations or miss key features that could dramatically affect a home’s value.
The Zillow effect: Does a low ‘Zestimate’ turn off homebuyers?
Zillow doesn’t hide the fact that its property-value estimates are just educated guesses. But some real estate brokers can find its outsize influence grating.
But algorithms play a growing role in all manner of business transactions, from the variable costs of Uber and Lyft rides to airlines’ dynamic pricing for seats.
In the real estate world, some companies — including Zillow, but also competitors like Redfin and Opendoor — are now offering to buy houses based on their algorithmic valuations.
It remains to be seen how many homeowners will risk leaving money on the table for the sake of convenience.
Zillow, for it’s part, says its computer-generated estimates are just a starting point that might differ from an actual appraisal.