Buying a home is a huge financial decision. And with any property, there are always unknowns. But new research finds AI will help give buyers and sellers more information to work with. While the process will help people across the socio-economic spectrum, buyers and sellers in lower-income neighbourhoods will see the greatest gains.

In research published in Marketing Science, Rotman professor of marketing Nitin Mehta and his co-authors found that the U.S.-based property value estimation tool Zestimate helped make the real estate market more efficient, with sellers who used the tools seeing between four and five per cent profits.  

Zestimate is a machine learning algorithm developed by Zillow, an online real estate marketplace similar to Canada’s Realtor.ca. Its database includes information on more than 100 million properties listed on the marketplace over the years. And Zestimate leverages this data to generate a public estimate of a property’s value based on government records and information provided by users, such as its location, square footage, number of interior rooms and appliances.

Looking to measure Zestimate’s impact on listening decisions, selling price, time on market, buying surplus (the difference between the selling price of a property and maximum price a buyer would have been willing to pay for it) and seller profit, the researchers looked at 4,027 properties listed on Zillow in Pittsburgh, Pennsylvania, between February and October 2019. 

The study found that because the Zestimate reduces uncertainty surrounding actual property values, it enables sellers to reduce the likelihood of selling their properties below market value. It also helps buyers to match with the homes they truly value, leading to purchases that better fit their specific needs and preferences. 

“Having more information just makes a market more efficient,” he says. “Zestimate helped people in poor, mid-range and rich neighbourhoods. But we were surprised to find that it helped people in poor neighbourhoods the most.” 

The study included properties in 140 neighbourhoods, which Mehta and his co-authors classified as poor (those with a Zestimate — Zillow's estimate of a home's market value — below US$130,000), mid-range (with a Zestimate between US$131,000 to US$279,000) and rich (with a Zestimate above US$280,000). 

While the study found that Zestimate was less accurate in lower-income neighborhoods, those same areas benefit the most because the tool helps reduce baseline uncertainty around property values, which is most common in these lower-income markets. 

Mehta attributes this uncertainty to the fact that many do not have access to high-quality realtors, and the market information about properties is skimpier in poor neighbourhoods. 

The study found that in poor neighbourhoods, a Zestimate increased seller profits by 4.78 per cent. In contrast, seller profit increased 4.19 per cent in mid-range neighbourhoods and 4.21 per cent in rich ones. There was an even bigger difference in buyer surplus. In poor neighbourhoods, it was 9.09 per cent. While in mid-range and rich neighbourhoods, buyer surplus was 3.26 per cent and 7.26 per cent, respectively. 

“These algorithms create a win-win environment for both seller and buyer, and they don’t need to be perfect to create real value,” says Mehta. “Markets respond strongly to any signal that reduces uncertainty, and Zestimate does exactly that.

“The belief has always been that AI would help more educated people and richer neighbourhoods — that it is not a tool for poor people,” Mehta adds. “Even though Zestimate’s performance is not as accurate for poor neighbourhoods, it actually helps poor neighbourhoods more.”