Zillow’s Zestimate is now 15 percent more accurate when it incorporates artificial intelligence that can determine how nice a home is, like a human would.
The Zestimate has always incorporated key factors to determine the value of a home: square footage, number of bedrooms, number of bathrooms and location. Those are known as “structured data.”
But the home value algorithm has also learned to analyze “unstructured data” that can help it estimate a higher value for a home with granite countertops and stainless steel appliances compared to a similar home with a formica kitchen.
Technically, the Zestimate can’t tell the difference between granite and formica. But the algorithm is being trained to analyze pixels in images and learn that certain pixels correlate with a higher price tag. The magazine Popular Science noted these features and their improvement to the Zestimate in a recent story about artificial intelligence. Zillow first started developing what it called a “neural network” to make these distinctions in 2016.
So far, Zillow has only introduced these capabilities in King County, where Zillow’s Seattle headquarters sits in Washington state. Zillow plans to roll out the feature more broadly but a spokesperson declined to say when or where.
Besides distinguishing between two kitchens, the new Zestimate could also analyze two homes on the same street with the same number of bedrooms and bathrooms. Now, if Zillow sees photos of both homes and one has been renovated and its next-door neighbor is a decade or two out of date, the Zestimate can reflect that in its price determination.
These are all things that any person looking at a house knows instinctively: which house looks nicer and more expensive. But the Zestimate hasn’t been able to parse those subtler differences until now.
“We’re now trying to teach computers that same sensibility that a person has,” Zillow Chief Analytics Officer Stan Humphries told Popular Science. “Teaching a computer to appreciate curb appeal is truly artificial intelligence.”