The New Industry Leader in AI Home Search? In Conversation With the Team Behind Homes AI

Metadata

• Author: Harvey Hancock, , March 5
• Full Title: The New Industry Leader in AI Home Search? In Conversation With the Team Behind Homes AI
• URL: https://www.onlinemarketplaces.com/articles/the-new-industry-leader-in-ai-home-search-in-conversation-with-the-team-behind-homes-ai/

Highlights

• CoStar Group’s ‘Homes AI’ has the potential to become the most impressive search tool to hit the real estate industry in years. I spoke to Andy “Ace” Ventura—head of Applied AI—and Livia Sponseller—Head of Product Design—about the birth, development, launch, and future of Homes AI. From conceptualising the product in November 2024, to its launch in February 2026, here’s what they had to say. (View Highlight)
• At their core, online marketplaces are a ripe environment for innovation, particularly in the highly emotional, high-stakes home search space. “It’s voyeuristic,” says Livia Sponseller. “It’s fun. We spend a lot of time fantasising about our next home. And AI is a much cooler way to search. We started conceptualising this feature around fifteen months ago, but we only really started pushing it in the past several months of focused work. We’ve come a pretty long way.” (View Highlight)
• Homes AI, now live on Homes.com as a speech-to-speech search experience, actively filters out unappealing listings, changes the images that consumers see without them needing to click a single button, and does so in “fifty” languages (English, Italian, Spanish, Russian, and Arabic have all been demoed or at least namedropped). The result is an interactive, conversational search experience that changes the website you see. Homes AI is highly refineable, it’s relevant, and takes browsers one step closer to narrowing down on their dream homes as more ideas spring to mind. In many ways, Homes AI rewards the curious and imaginative consumer, equipping them with as much knowledge as they’re creative enough test the AI on. (View Highlight)
• As Andy Ventura describes it, Homes AI is a major breakaway from traditional filtered search. Browsing for a property no longer needs to be defined as a “click, click, scroll” experience. Speech-to-speech was a logical next step. An ongoing partnership with a small company called Microsoft was key. (View Highlight)
• The first problem was to make it easier to search and identify properties, says Ventura. Something less tag-focused, and more akin to an open-ended Google search. (View Highlight)
• “We spent a lot of time building all these great filters, and we were seeing that people were using the basics like beds and baths, and that was about it. We felt like we wanted to unlock the true power of our search engine. (View Highlight)
• “It took a long time for us to get right. One, doing natural language processing for search, with models like ChatGPT and Claude, is slow. Two, it’s expensive, and three, it’s not that accurate in edge case scenarios. We had to build a custom model that handled [every prompt]. ChatGPT was a big inspiration, how interactions suddenly felt like magic. We realised that with all the proprietary data that we create at CoStar Group about neighbourhoods, home price trends, sales comparisons, Matterport, that’s when we knew we had the tools to create an experience that feels like it’s from the future. (View Highlight)
• “Microsoft has a very good security model in terms of making sure the model can’t misbehave, and that the data stays with us. That was a requirement we had from early on. When we showed them a prototype [of their speech-to-speech API combined with CoStar Group’s proprietary data] they were like, ‘wow!’ They were used to seeing clients build customer service chat bots, that was the only use case they were getting. It’s a valid use case, but we wanted something [much] bigger.” (View Highlight)
• The technicalities of building a fully responsive speech model deeply embedded in the architecture of a property portal are, frankly, too complicated to be entertaining for our readers. The short version is that it took a lot of manpower, brainpower, investment dollars, and time to develop an AI that could be trusted to give relevant, informative answers to user queries. (View Highlight)
• Sponseller suggests that a blended experience, which can interact seamlessly with the frontend of Homes.com’s portal, is a key differentiator. A siloed AI search that takes consumers away from the core web experience they’re used to simply wasn’t an option.

  “We blended the AI into the site architecture to bring out the ‘magical’ experience for a consumer who is mostly used to seeing a chatbot slapped onto the bottom right-hand corner of their webpage. These chatbots don’t do any harm, and they’re a valid use case, but that wasn’t our ‘north star’ for what innovation should look like in this space. OpenAI’s chat app is nice, but it’s very limited in what it can do.” (View Highlight)

• The result was an AI that closely resembled the intuitiveness of the user interfaces that consumers are used to with the handheld devices they use every day. “The AI and the UI are in the know about each other,” says Ventura.

  “You still have a mouse and keyboard, you still have a very powerful touchphone on your iPhone or Android device. If you’re talking one moment and clicking the next, the AI needs to stay up to speed. [If our user was] on the summary page, but all of a sudden they’ve opened up the photo gallery and are looking at kitchens, the AI needs to know what you are seeing. [This combination is] a big part of what we felt like was going to differentiate us from chatbots or embedded applications in ChatGPT.” (View Highlight)

• Ventura suggests several improvements to Homes AI are already in the pipeline, including better memory (the AI will remember you want a home with a garage, for example). More data, including exposure to natural light and proximity to grocery stores (and their opening hours), are also namedropped without being officially committed to. Sponseller noted that AI will be learning certain characteristics and give contextual search results in the near future. For example, Homes AI may direct you towards a particular listing if it knows that you like a spacious kitchen or a sunny master bedroom. (View Highlight)
• There’s so much to be excited about with Homes AI. It’s an innovation that genuinely feels like several important steps forward for a notoriously slow-moving industry. Yet the annoying truth of bringing a brand-new product to market is that teething issues are inevitable—and they also need to be actively planned for. Which brings us to the bad news—sometimes, clicking is cleaner. After all, clicks can never go wrong in quite the same way as an AI search. (View Highlight)
• Lagginess is a problem, if not a dealbreaker, while some personal testing in my free time threw up an annoying prevalence of the AI listening to a (fairly simple) request and then ignoring it, or asking for it to be repeated…and then ignoring it! Meanwhile, Homes AI threw some factually incorrect data into Online Marketplaces Chairman, Simon Baker’s (niche) home search in Miami, while also missing out on some of the nuance that goes into searching particular edge case search parameters. (View Highlight)
• Homes AI is a bold and admirable product, but for now, the de facto ‘best’ solution to AI search feels clunky at times. As Online Marketplaces editor, Edmund Keith, suggested on a recent episode of the PPW Podcast, Homes AI feels like a product that was launched too soon. (View Highlight)
• Sponseller rightly points out that the best learning can only come from the product being available to the public. Her gut instinct is that optimising the AI for paying, member agents will be a high priority as the product evolves as part of the customer journey. (View Highlight)
• “We want Homes AI to be a perk, not a competitor. Letting [agents] train the AI a bit before they show their consumers [is something to focus on]. We want professionals to have agency on where the Matterport tours start, and how to ensure the AI isn’t misleading consumers into making offers without an agent’s help. Every transaction requires a good agent. (View Highlight)
• “AIs are cool, but there are some things they can’t help with. Maybe we have to make the agent search a little bit more AI-friendly; that’s on our radar as well. When a consumer is ready to make an offer, we want the AI to help them find the best agent for them, someone who is familiar with the area.” (View Highlight)
• Ventura agrees there’s work to be done.

  “Giving users control over the AI. Take chattiness, for example. Some people want an active AI. Others want a ‘speak when spoken to’ approach. We’re at the bleeding edge here. Microsoft told us we are ‘one of one’, so there are a lot of boundaries we need to take into account, like transitioning from one screen to another, we’re not fast enough yet, and the AI isn’t enabled for the whole search yet.” (View Highlight)

• From an internal perspective, there’s another problem that bears thinking about. One thing that becomes abundantly clear when discussing the technicalities of Homes AI is that it must be eye-wateringly expensive to operate. CoStar Group appears to be highly aware of this and was understandably reticent to share numbers. (View Highlight)
• Ventura says the team had to work hard to minimise the cost of each voice search, using a mix of several models and reducing its reliance on complex logic chains to keep things simple and lean wherever possible. One benefit is that once a particular question has been answered once by the model, it’s faster and cheaper to retrieve that information when it is requested, rather than calculate the same answer afresh. (View Highlight)
• Homes AI is also good at saying no. It will reject generic requests from consumers who dare to ask it which football team they should support, or recipes suitable for big dinner parties. “I can’t give away our secrets here too much,” says Ventura, “but once you go ahead and get a model to do something very specific, you can go from a very long set of instructions, paragraphs long, to a very short set of instructions that is an order of magnitude less in cost.” (View Highlight)
• He mentions that the first version of the Speech API was much more expensive than the one that went to production. “We have controls in there to make sure it doesn’t go off the rails.” The final complication is a big one: how does a portal like Homes.com market such a drastically different, technology-driven search that most consumers simply aren’t educated on? Sponseller admits that marketing a new solution is an “iteration game” because silver bullets don’t exist yet.

  “You’re hitting on what is probably becoming the most difficult part of this process. It’s hard to boil Homes AI down to something digestible when everyone’s seeing thousands of ads every day. We’re going to quickly pivot and iterate. Simpler is better.” (View Highlight)

• She suggests that Zillow enjoyed a major win when it launched its ‘Zestimate’ product, allowing consumers to get a realistic home valuation for the first time.

  “Everyone flocked to Zestimate. Zillow is our biggest competitor, but I respect what they’ve done. If Homes.com can have a similar ‘Zestimate’ moment where consumers are enjoying their home search experience, that would be a fun way to get some brand awareness and adoption.

  “Success would be helping people find their ‘needle in a haystack home’ faster and more efficiently.” (View Highlight)

• This appears to be a realistic target. Several users during user testing were won over by an intuitive interface and the injection of ‘fun’ into a home search experience. If fun and intuitiveness bring people back for more, “Great. Ecstatic,” says Sponseller. (View Highlight)
• Homes AI will be rolled out for other businesses in the CoStar Group network, too. Loopnet, Land.com, CoStar and Apartments.com are all namechecked, among others. “Apartments.com is next,” says Sponseller. Meanwhile, Domain’s rollup into Homes.com suggests that it is only a matter of time before Homes AI is launched in Australia. Creating connectivity across a family of marketplaces will create a new ecosystem for portals, according to CoStar. (View Highlight)
• Ventura sees the consumer/agent relationship as his metric for success. He imagines a scenario where an agent feels empowered to send their clients to Homes.com to search, and then come back to the agent when they’re ready to visit a property they’re serious about buying.

  “I think my definition of success is this: that our customers, our member agents, are so blown away by Homes AI, that they say, ‘This is gonna save me time’. When they’re meeting with clients for the first time, they’ll say, ‘Go do your home searches on Homes.com. Homes AI will answer all your questions, and when you’re ready, we’ll go see this house, and I’ll get you the best deal I possibly can’.” (View Highlight)

• I. When it comes to real estate and AI search, what isn’t the industry talking about that it ought to spend more time talking about? On this, both Ventura and Sponseller agree: the role of the agent needs to evolve, and fast. “Agents are very specialised,” says Ventura. “They know the areas. They know the contractors. They know the appraisers. But they may not know the investment value of a property; they’re not financial experts. I think agents should be using these models to help them help consumers make better decisions with information that may not be in their traditional purview.” (View Highlight)
• Sponseller says the elephant in the room is that AI closes the knowledge gap between consumer and agent—so agents will need to work harder to stay ahead of the game.

  “[With AI] you’re no longer just a consumer on a home search site, looking at a listing. You’re accessing insight. You are getting a 360 view: home values and sale history, school rankings, who’s the principal at this school? AI democratises information that used to be siloed for agents. Consumers are now more empowered and better informed.” (View Highlight)