It’s a cool little internet widget that allows you to gather intel on your neighbors, watch your nest egg grow, and imagine moving up to your actual dream house. According to its creators, it’s now more accurate than ever.
An updated version of Zillow’s “Zestimate,” a tool that purports to show the value of nearly any American property, was launched Thursday. The company says the Zestimate can now “see” unique home features and pull more real-time information into its calculations, and claims an “error rate” of less than 2%.
The Zestimate, which first launched in 2006, was revolutionary in many ways. It turned Zillow /zigman2/quotes/205077794/composite ZG -0.17% /zigman2/quotes/204413973/composite Z +0.05% into the premier destination for ordinary consumers interested in real estate information, and helped launch the young media company into a destination for real estate agents who wanted to advertise to prospective buyers and sellers.
It was also a major milestone on the journey toward replacing human appraisers with what’s known as an “automated valuation model.” Now every source of information about real estate has its own for consumers, and the industry is moving toward allowing such models for things like mortgage approvals.
The Zestimate has become such an entrenched part of the residential real estate industry that some analysts now refer to new developments, like the rise of instant offers, a pivot Zillow is now executing as “Zestimate 2.0”
But Zillow’s 2.0 is here today because the first incarnation had limitations. Consumers complained – though they still clicked – and some even sued. So in 2017, Zillow launched a contest to encourage data scientists to come up with a better model. Takeaways from that contest are incorporated into the new version.
What’s new? Among other things, the model now pulls data from homes currently listed for sale, such as pricing and how long a property has been on the market.
It also uses what Zillow calls “neural networks and computer vision” to “see,” in the company’s words. “While the Zestimate can’t interpret home features by name, the way a human would, the image recognition model can classify patterns in the pixels of photographs and correlate them to home value,” the company said. “For example, while the human eye sees tile or granite countertops, the Zestimate identifies two different pixel patterns.”
And about that “error rate:” Zillow says its boast of “less than 2%” means that “half of all Zestimates fall within 2% of the home’s eventual sale price.” Half don’t, of course — and some may continue to fall very wide of that target.