keys beside miniature house
An arrow pointing leftHome

Hey, Househunters: AI Is Here to Help

  • Alexander Gary
12/20/2021

From finding a property to securing a mortgage, companies that assist in the homebuying process are increasingly leaning on algorithms to place people into their forever home — despite what happened to Zillow.

Early in November, real estate company Zillow was forced to drop the curtain on its Zillow Offers program: an algorithm-based service that allowed the real estate behemoth to purchase homes from sellers for quick cash, then flip those properties after a tidy clean-up. This system, known in the housing market as iBuying (a.k.a. “instant buying”), remains in practice by many other real estate agencies — Offerpad, Redfin, Opendoor — and is meant to step over all the time and hassles selling a place would normally entail. And while Zillow didn’t do everything right, some other players, from money lenders to appraisers, continue to succeed.

Modern real estate companies such as Divvy Homes, Compass, and even Realtor.com are applying the magic of AI algorithms to help streamline the process for buyers on one side, sellers on the other, and all the brokers and agents in between.

Say you were looking to buy a home a decade ago. After running through the necessary channels (such as government and private offices offering information on renovations, square footage, property values, etc.), you would then have to secure pre-approval of a mortgage to know what you could afford, then find a broker to help you isolate a neighborhood and specific homes to find the ideal location and space. It all feels a bit old-fashioned now. No longer.

Today, buying a home can be easier and more equitable than in decades past, thanks to brokers and banks who employ AI platforms to do everything from streamlining the home-selection process, to helping assess market conditions, to finding the right fit for a mortgage.

Akshat Kaul, head of machine learning for Seattle’s own Redfin, says that his company averages nearly 49 million monthly users, and that they are regularly engaging with Redfin Estimate — an AI calculation of the market value of a given home — as well as Redfin’s home recommendations tool. “The Redfin Estimate…in combination with the expertise of our real estate agents, can help our customers find a very good strategy for either bidding on or listing a home. Similarly, our home recommendations tool [was developed] to recommend homes to people based on other homes they [should] be looking at on Redfin.com,” he told us about his firm’s AI systems. “Almost a quarter of all sessions on Redfin are driven by people looking at homes that we recommend to them, and that’s up from 8% just five years ago.”

In this way, AI technology helps to move the market forward by learning more about users and their desires based on their behaviors and relying on analytics and scheduling to work with realtors in an environment that demands fast reaction times.

(And, hey, even if you have no intention of buying, looking is perfectly natural. Just don’t do it with your kids in the room.)

For sellers, AI platforms can be the deciding factor between constantly looking and narrowing down the perfect buyer. Restb.ai, an AI company that specializes in real estate, helps lookie-loos sift through photos by using computer vision models to automatically identify parts of a home with special tags, as opposed to the tedious, endless photo-clicking anyone who bought a home in the past decade is used to.

Seattle-based CityBldr’s locates land in more than 100 cities across the U.S. and software analyzes its potential capital by tracking the latest data market and zoning rules. Then, the software assists sellers by adding a seller’s property to its database of hundreds of builders, developers, and investors, and promotes that property by using buyer prediction tech that zooms in on whoever may be looking to invest. It creates an instant marketplace.

Real estate agents and brokers in the middle are also able to use AI platforms at their disposal. For example, Zillow still uses natural language processing to track what shoppers may have said and wrote about properties they may have seen with Zillow’s representatives. As Zillow’s estimates are not technically appraisals,the company balances its numbers by using NLP to gather the complete context for what a customer wants to know to help refine his or her search. Other new companies are easing their way into determining the true value of a property. Quantarium, another Seattle-based AI company, combines the knowhow of software, data, and real estate experts to create a system that determines property valuations and predicts prospective customers. Quantarium’s machine learning platform determines more than one billion property values monthly and is used as a resource by home construction companies, marketing agencies, mortgage lenders, and financial institutions across the country to help figure out what a property should sell for. One of Quantarium’s clients last year was Realtor.com, which had also partnered with Collateral Analytics and CoreLogic to “[make] buying and selling homes simpler and more enjoyable for everyone,” as the company’s Vice President, Todd Callow, put it. It has since added a three-home-value estimate function to give people a fuller picture of what the value of a property might likely be. The theory is that if people only see a single estimate, they may create misinformed ideas based on one number. While there is no one-track method of buying and selling a home, Realtor.com hopes that Quantarium’s AI solution can help everyone to understand that other options do exist.

AI programs are also helping to reshape the mortgage industry. Much of the industry still relies on paper documents and keen human eyes to fill out documentation or catch mistakes. AI programs save both time and money by reducing the use of paper, tracking borrowers (specifically, whatever patterns they may display before they make the move to buy a home), moving data between systems, and gathering information on the market in real-time to better service clients. Dotloop, an online platform that serves buyers, sellers, and realtors in the transaction process, has made a complete switch from physical paper to technology to better organize files and collect eSignatures (“a game changer for sure,” they note). Meanwhile, Zest AI, a Los Angeles software company, is attempting to improve the credit scores of a wide range of Americans by using machine learning to help them find lenders. Its operating system relies on alternative data gathered from other aspects of a borrower’s life that may not necessarily be counted toward a credit score — like regular rent or utility payments — and adds this new information into a borrower’s approval ratings. (Just be mindful of the details; for example, not every credit scoring system takes rent payments into account, and the same can go for utility payments. But they can find their way into a report if they’re delinquent. Zest has claimed that its Black and Latinx clients — groups of people historically disenfranchised by credit and lending disparities — have had their approval rates raised by as much as 30% with the inclusion of alternative data.

Companies are even doing their part to better counteract the racial, gender, and cultural biases that AI programs unfortunately pick up, such as known cases of rejection for mortgage applications for people of color compared to White mortgage applicants. Redfin’s Kaul noted that machine learning algorithms get smarter based on whatever data is fed to them, so if the data itself is biased, then, naturally, the outcome of the model will reflect that. Redfin, he says, “[is] very careful and intentional about the data we feed into those algorithms. For example, we don’t add crime data as an input to our Estimate or our Recommendations algorithm because this data might create some biases in the model.” Even if customers may not agree, Redfin has taken this stance to better protect its customers. “It’s our responsibility to build algorithms that are equitable and fair and don’t reencode the same human biases that have led to poor outcomes, like neighborhood segregation. We belong in this society, and we have to care about it,” Kaul says.

With all the buzz around AI and real estate in the past decade, Zillow Offers seemed to be the logical next step for the industry as it increasingly relied on AI to carry the workload. The practice started in 2018, and was an attempt to step out from the position of the “middle-man” of a deal to the actual dealmaker. Three years later, Zillow had unfortunately lost $380 million (according to its Q3 reports) as a result of unforeseen mistakes in its iBuying approach, and was forced to lay off 2,000 of its employees in the wake of the event. Shortly after, it announced that it would sell 2,000 homes across 20 American markets to New York City-based Pretium Partners.

What, exactly, may have been the errors Zillow hadn’t accounted for that would lead to such a setback? Jeremy Wacksman, Chief Operating Officer, had said that they were “operating within a labor-and-supply-constrained economy inside a competitive real estate market, especially in the construction, renovation, and closing spaces.” Or, to put it less tactfully, the economy hasn’t been working in Zillow’s favor: wages in the United States are rising, but not nearly as fast as home prices. The price of building supplies is high, while the actual construction of new homes is in decline, coinciding with a slower release pace for the building permits of fresh units.

It’s certainly possible that all these elements working together is what led to Zillow Offers’ premature casualty, but this explanation doesn’t consider any of the technological hiccups Zillow may have also faced. Zillow Offers was based on machine learning, starting with “Zestimates,” Zillow’s personal estimate of a home’s market value that relies on data the company retrieved from public records and personal submissions by users (such as detailed photos of rooms, tax assessments, or the value of other homes in the area). Taking everything into account, Zillow, then, stepped in and made its offer. As Zillow quotes on its own website, its purchasers “use state of the art statistical and machine learning models that can examine hundreds of data points for each individual home.”

What may have been Zillow Offers’ fatal flaw was an overreliance on the AI program that was meant to work with Zillow’s employees, not handle all the heavy lifting. Zillow’s algorithms had trouble reading the market as the pandemic continued: in February, Zillow’s sales outperformed quarterly projections, as people were seeking new homes, likely due to COVID fatigue and a desire to leave certain hot zones. Zillow, then, was able to buy and sell as many properties as it could fund. But when the “labor-and-supply-constrained economy” showed its face and the market began to slow, Zillow was left with more homes than it could make a profit from. Zillow Offers’ needed to adjust its algorithms to account for changes in the industry, but instead, took the risk of letting the algorithms do the work for the company. As Zillow spokesperson Viet Shelton admitted to CNN Business, the company’s problem was an inability to predict for the prices of their inventory a few months ahead of time, in a market where changes were happening fast.

This isn’t a reflection of the dangers of AI. Rather, Zillow’s iBuying failure should serve as a wake-up call to those who believe AI platforms are meant to replace human savviness. We should be working with our AI programs and understand that they have just as much to learn from us as we do from them. After all, AI systems can collect data that confirm a house has an issue with its foundation, but it needs an agent and a seller to see these things first before it can report it. AI can’t predict frequent flooding if it doesn’t know the issue is even there.

Other real estate companies could learn from this episode. In the wake of Zillow Offers’ demise, Redfin’s CEO Glenn Kelman has fully acknowledged that they, as iBuyers, took the slow and steady route of both making fewer offers and paying less for every home they came across. And as Kaul mentioned, Redfin’s AI works with the expertise of its agents, not despite it.

The real estate market may be becoming more science than art, but the artists behind this science should always remember how important they are to the show. “We’re trying to solve the big problem that real estate customers [face], which is to navigate the complex real estate transaction,” Kaul says.