Zillow wants to reinvent the home-buying experience using AI. Leveraging Zestimate and other new tools and acquisitions, the property platform is aiming to become a one-stop shop for renting and buying a home. But just how central is the Zestimate to Zillow’s strategy, and has it helped or hurt the company’s mission?
Category
🤖
TechTranscript
00:00This is the biggest financial transaction that most people will take in America in their lives.
00:05Zillow is betting on AI to simplify the path to home ownership.
00:09Well, I'll tell you what's one of the most stressful things I've ever had to do, and that's buying a house.
00:13The real estate platform wants to be a go-to for all parts of the home-buying process.
00:18And key to this is its signature Zestimate.
00:20I'd be hard-pressed to find someone who doesn't know what a Zestimate is.
00:23Amid criticism of the tool's accuracy in its use of AI,
00:26will Zestimate and the company's use of AI tools help it maintain its dominance in the property tech market?
00:36So I think when a lot of people think about technology and Zillow,
00:40a first thought is Zestimate.
00:42Yes.
00:43Can you explain, just a bit to start, how that technology has changed over time?
00:49Zestimate is one of the first machine learning algorithms we ever put out at Zillow.
00:54It used to be this family of very interconnected, complicated models.
00:58And now, at the center, it's a neural architecture.
01:02And what that enables for us is the Zestimate now looks at both structured information,
01:08like the location, the beds, the bathrooms, as well as unstructured information.
01:13Think things like the images, the 3D walkthrough,
01:16to get that deep sense of the quality of the home and the value of the home.
01:20And that's really improved the accuracy of the Zestimate.
01:23Zestimate uses data points like tax assessments and sales transactions
01:27to provide a natural starting point for homebuyers.
01:30I think the Zestimate's the line in the sand.
01:31Everyone's saying, which side are we going to end up landing on when this transaction closes?
01:35If you really kind of look at the Zestimate, it's very good at kind of saying,
01:38hey, this is an estimate, this is kind of a potential range outcome,
01:41because a trick with residential real estate is what has kind of been done to the inside.
01:46We place Zillow as kind of a category winner who's been taking share
01:50and is essentially a ubiquitous dominant brand within a category,
01:53which is residential real estate.
01:55Competitively speaking, really no one's near Zillow in terms of traffic volume,
02:00the leads they can generate.
02:01Zillow's traffic volume is almost double compared to homes.com,
02:05with Zillow reporting an average of 217 million monthly visitors
02:09and homes.com's reported 110 million monthly visitors.
02:13But Zestimate has also faced criticism,
02:16frequently fluctuating home values, a lack of human touch in price predictions,
02:20and a failed home buying program.
02:22And Zillow is currently edged out by competitors on price accuracy.
02:25Redfin reports their estimate has a 1.99% median error rate for on-market homes,
02:32compared to Zillow's 2.4%.
02:34For them, if it's problematic, it's actually a much bigger risk to their brand.
02:39If there's a problem with this estimate, it's really distorting folks' expectations.
02:43The fact that their traffic volume is still leading,
02:46is still dominant in the industry, and it's enabled them to add more tools,
02:50would suggest that the Zestimate is a net positive for the company and consumers.
02:54You know, as a publisher of content, they understand if you put bad things out there,
02:59it can absolutely destroy your brand.
03:02And all the data would suggest that they've done a very good job at making sure folks
03:06really understand what it is, and it's a starting point.
03:08Some question whether tools like Zestimate take data at face value.
03:11Ignoring local histories, these concerns are most acute in gentrifying neighborhoods,
03:16where rising property values can overshadow the lived experiences of longtime residents.
03:21Zillow emphasizes that a critical aspect of building their technology involves
03:25programming it to be fair and equitable for all users.
03:28We make sure that people of different socioeconomic backgrounds or racial backgrounds
03:33aren't getting a disparate result.
03:35We're doing a lot with large language models.
03:38And in real estate, agents or digital experiences can't answer certain questions.
03:43We actually had to build a model that detects those kinds of questions and cuts them off.
03:49And we put that out as open source this last summer.
03:52Zillow's investments in AI tech include a $400 million cash acquisition
03:57of a customer relationship management system, a virtual staging company,
04:01and a media software and content management platform.
04:04The company has invested over $4 billion in tech and product innovations over the past decade.
04:10I think AI can do a lot of the stuff that you just don't need to be involved in.
04:14Note taking, drafting what home to look at, seeing if it's available for a tour.
04:18But there's still that agent, buyer, human element of going,
04:22hold on, I have unique needs, unique wants.
04:25They've changed over time.
04:27They really focus on a strategy of a super app.
04:30They are really creating a platform where you can find a house to buy.
04:33If you're not ready to buy, you can find an apartment to rent.
04:36And maybe you're renting and now you're kind of between both.
04:39They're helping you figure out what you can afford.
04:41And they can help underwrite mortgages or connect you to a mortgage broker.
04:44So they are solving all of the pieces for someone who essentially needs to figure out where to live.
04:48For Zillow, accuracy and customer trust is an important part of creating its AI tech.
04:53We want to make sure people are confident in the answers they're getting from us,
04:57be it, you know, the price or beds and bathroom stats or something that AI is generating.
05:01And so we do a lot of testing to make sure people understand and know how to interact.
05:05And we find that most people already expect AI is behind the scenes.