Rocket Homes Tech Chief Unpacks ‘Explore Spaces’ Computer Vision Tool
At Inman Connect Las Vegas, July 30-Aug. 1, 2024, the noise and misinformation will be banished, all your big questions will be answered, and new business opportunities will be revealed. Join us.
As one of the nation’s largest mortgage lenders, it’s no surprise that Rocket Mortgage has embraced generative AI, automation and machine learning to gather and process documents needed to underwrite borrowers and close loans faster.
But it may come as news to real estate agents that Rocket Mortgage’s sister company, Rocket Homes, is employing AI to revolutionize the way consumers look for homes.
RocketHomes.com’s new visual search tool, Explore Spaces, uses an AI capability known as computer vision to recognize and process listing photos, generating text descriptions so images don’t have to be tagged manually.
TAKE THE INMAN INTEL INDEX SURVEY FOR MAY
The idea is to enable users to discover their dream home by sifting through mountains of listing photos to pull those that display features that are important to them, like kitchens with marble countertops or backyards with lush lawns.
RocketHomes.com, which averaged 1.5 million monthly unique visitors in 2023, licenses listing data from about 200 multiple listing services nationwide — including photos uploaded by agents.
Visitors who use Explore Spaces to search for listings spend almost twice as long on the site per visit, and return six times more often, Rocket Companies CEO Varun Krishna said on the company’s May 2 earnings call.
Rocket Homes Chief Technology Officer John Fair recently shared some insights into how Explore Spaces came to be, and how real estate agents can make the most of the new approach to search.
One key takeaway: While professional photos help buyers see a home in the best possible light, the quantity of images is also key to catching the eye of AI and showing up in visual searches.
Inman: Visual search is a departure from the traditional filter and keyword approach employed by most real estate search sites, but it sounds like RocketHomes.com is seeing significantly better engagement.
As we’ve been watching what’s happening in the industry, how others are leveraging AI, we sort of saw a space here for us that we’ve named Explore Spaces, which is intended to help the typical homeseeker visualize living in a space, or focus on kind of the key criteria for their dream home. The idea behind Explores Spaces is most people are probably searching visually for their ideal home. They want square footage, they want bedrooms, they want bathrooms — those are typical filters.
But also, NAR says about 84 percent of buyers believe that photos are the most valuable feature on any home search site. We take that and we kind of synthesize it into this idea of: How do we bring photos to the forefront for the client experience? How do we really make them the focal point for a search for a client while they’re searching for a home?
So as we’ve looked at that, as the primary feature of Rocket Homes Explore Spaces, we have seen some increases in engagement. Things that we believe over here at Rocket Homes are really going to make a difference in ‘Helping everyone home.’”
I’m guessing that you can’t do this without AI — that’s a critical component of this.
We’ve leveraged things like computer vision and AI to look into each of the photos for each listing. So we take these photos, we’re weaving AI in to look at them through the lens of what’s in this photo? Is this a pool? Is it a kitchen? Is it a bedroom?
We use that to provide to our clients who’ve identified to us, “Okay, I’m really interested in kitchens, and I’m looking in, let’s say, Phoenix, Arizona, I’m really focused on and my wife wants a white kitchen” — bringing that to the experience for the consumer so they don’t have to go through every listing. You know, maybe that’s brown cabinetry, or it’s primarily mahogany, they’re like, “What I really want is a white kitchen,” bringing that to the forefront.
So that once a user sees the kitchen that they’ve always dreamed of, they can click on that image and it takes them directly to that listing, kind of simplifying and easing the search process by using AI.
So computer vision helps you interpret the information that’s in the images uploaded with listings, but how do you learn what the consumer’s ideal kitchen is? Are they kind of training it by what they click on? How do you get that information into the system?
The name of the game here is user behavior — it’s interactions based. What we’re doing, is we’re using what they’re spending time on, interacting with, so if they’re spending more time seeking out, you know, that white kitchen, what we can do is make sure we’re surfacing more photos that are akin to or similar to what they’ve interacted with.
So the more you’re using it, the better job it’s going to do of surfacing stuff that you like?
Yeah, that’s obviously the intent of most AI algorithms, to use what you know about an individual client’s intent or their interaction with the product to improve long term. So that’s obviously part of our roadmap of things that we want to improve upon, because Explore Spaces just launched here, probably five, six weeks ago now. So this is stuff that we’re constantly iterating on and improving upon.
So did Rocket Homes develop this in-house? Are there folks that you’re working with that have helped you develop this?
I think with any technology, you’re always looking for best in breed. So while we don’t really get into how all of the algorithms are being created and where they’re sourced from, I would say we’re looking at all potential partners out in the universe.
We’re looking at what we’re building internally, and we’re really seeking the best in breed because our focus at Rocket, but also Rocket Homes, is ensuring that the client experience is the first thing.
We’re focused on AI. We want to make sure that our client experience is top of mind. It’s the first thing we’re thinking about whenever we’re developing or thinking about a new feature like Explore Spaces.
From the user standpoint, is visual search just something that happens automatically if I’m looking for homes on RocketHomes.com, or do I have to opt in to this feature?
You can choose to interact with Explore Spaces or not. Our typical home search experience is still available to you. And then if you want to use Explore Spaces, you’ll notice options on the right-hand side to choose. Say there’s a drop-down, like, “I want to see kitchens,” and that will actually change the user interface so that the primary photo shown for each listing is kitchens.
We’re not making the choice for the consumer. I think like with anything out there, we’re always running tests to see what is piquing consumer behavior, what’s most interesting to clients and making sure that we’re optimizing again for that client experience. That is always at the forefront.
You’re able to do this in all markets, or is it just in limited markets right now?
It’s in all markets right now. Of course, that means, you know, up here in Michigan, pools aren’t as ubiquitous in Michigan as Phoenix.
So there are going to be some places where you know, maybe pool, if it’s what you’re looking for, it’s not as common in North Dakota, maybe, or in colder climates than it is in warm climates.
So there’ll be specific features that wouldn’t be as applicable, but you can use [Explore Spaces] in any market across all 50 states right now.
So from the listing agent’s perspective, does this mean I have to be thinking differently about what kind of imagery — and how much imagery — I upload to the MLS?
I think Realtors in general are already super aware that photos sell homes, right? So I don’t think that would necessarily change Realtor behavior here because they’re already very focused on taking high-quality photos. So I don’t think that it’s added any additional burden for real estate agents so much as they can benefit from this feature helping them highlight the cool features of a property that they’ve listed.
The good part about the way that we implemented Explore Spaces is it’s not necessarily as dependent on the quality of the images so much as how many images we have. So the more the better. Regardless of the price point of the listing, our product still operates. So it brings value to the market.
I think that really speaks to our vision here — of streamlining what is typically a very complex process, both home search but also home purchase. So that really opens up here, for real estate agents to not have to have as much of a focus on maybe the [quality of the image] as the scope or the number of images that they have across all rooms and all aspects and facets of a property.
It’s really helping them make their listing that much more accessible to whichever homeseeker will ultimately end up in the property.
Get Inman’s Mortgage Brief Newsletter delivered right to your inbox. A weekly roundup of all the biggest news in the world of mortgages and closings delivered every Wednesday. Click here to subscribe.