Editorial illustration for Zillow's AI strategy: Build before measuring ROI, own the chat layer
Zillow's AI Strategy: Own the Chat Layer First
Zillow doesn't get a tidy, single conversation with the people who use it. Someone browses listings on their phone in March, talks to a loan officer in June, and sits down with a real estate agent in August, and Zillow expects to remember all of it. That's the problem Toby Roberts, Zillow's SVP of Engineering, laid out at VB Transform 2026, alongside Glean co-founder and CEO Arvind Jain.
Zillow's products run through roughly 80% of U.S. real estate transactions each year, and the company has been building AI systems since long before ChatGPT made the term fashionable. Scale like that means a chatbot bolted onto a website was never going to cut it.
Roberts said the team knew early on it needed something that could follow a customer and the professionals working with them across every surface they touched, not just answer questions in isolation. That meant rethinking what "AI infrastructure" even means for a company whose customers disappear and reappear months later expecting to be known. Roberts and Jain's conversation traced how that realization reshaped Zillow's approach, starting with a problem everyone assumes is the hard one: the data itself.
Build the measurement baseline before the AI push, not after. Roberts said Zillow's ability to credibly attribute a 40% increase in shipped code to AI adoption rests on a DORA metrics baseline the team put in place years earlier, not on the AI rollout itself.
Why this matters
Roberts' framing is a useful corrective for anyone building AI products right now: ROI math done after deployment is mostly theater, because you've already locked in your architecture by then. The real decision point is earlier, when you're still deciding whether a single chat interface can actually hold your customer's journey. Zillow's case is instructive precisely because its transaction isn't a chat, it's a months-long relay between a phone screen, a loan officer, and an agent, and no off-the-shelf chatbot was going to carry that context without dropping it somewhere.
That's why Zillow built its own layer instead of outsourcing the interface to a vendor, using Glean as infrastructure rather than the front door. For founders and engineering leads, the lesson isn't "build everything in-house." It's that owning the context layer is a strategic choice tied to how fragmented your customer journey actually is, not a default. If your product involves handoffs across people, time, and channels, the question to ask before shipping anything is whether a single conversational interface can survive that handoff at all.
Related Reading
Business & Startups Nordic pilot adds Gemini for Education, NotebookLM to boost AI literacy
Business & Startups Kling launches Video O1, all-in-one model with MVL bridge using transformer
Business & Startups DeepSeek Seeks More Capital Weeks After USD 7B Funding Round
From Archives Monitoring Beats Testing for Catching AI Failures, Experts Say
From Archives Bristol Myers Squibb Builds SuperDuperPOD on NVIDIA Vera Rubin
Common Questions Answered
What is Zillow's main challenge in implementing AI across its real estate platform?
Zillow faces the challenge of creating a unified AI experience across fragmented customer journeys that span multiple touchpoints over months. Users browse listings on their phone in March, interact with loan officers in June, and meet with real estate agents in August, requiring Zillow's AI to maintain context and memory across all these interactions. The company needs to own the chat layer to successfully integrate these disparate experiences into a cohesive customer journey.
Why does Toby Roberts recommend establishing DORA metrics before implementing AI?
Roberts emphasizes that measurement baselines must be established before the AI push, not after, because ROI calculations done after deployment lack credibility and context. Zillow was able to credibly attribute a 40% increase in shipped code to AI adoption specifically because they had implemented DORA metrics years earlier, providing a proper baseline for comparison. Without pre-existing metrics, it becomes impossible to accurately measure AI's true impact on engineering productivity and business outcomes.
How does Zillow's scale in U.S. real estate transactions impact its AI strategy?
Zillow's products run through approximately 80% of U.S. real estate transactions annually, making the company's AI implementation decisions critically important for the entire industry. This massive scale means that Zillow's approach to building AI capabilities and owning the chat layer will likely influence how other real estate platforms approach similar challenges. The company's ability to successfully integrate AI across such a large transaction volume demonstrates the viability of the strategy for other large-scale platforms.
What does Roberts mean by 'owning the chat layer' in Zillow's AI strategy?
Owning the chat layer means Zillow is building a unified conversational interface that can connect and contextualize customer interactions across multiple channels and time periods. Rather than relying on separate chat systems for different parts of the customer journey, Zillow is creating a single AI-powered chat interface that maintains memory and context throughout the entire real estate transaction process. This approach allows the company to provide a seamless experience where the AI remembers previous interactions whether they occurred on mobile, with loan officers, or with agents.
Why is establishing architecture decisions before AI deployment critical according to the article?
Roberts argues that ROI math done after deployment is mostly theater because architectural decisions have already been locked in by the time you measure results. The real decision point occurs earlier when determining whether a single chat interface can actually support the customer's entire journey, not after the AI system is already built and deployed. By measuring and planning before building, companies can make informed architectural choices that will actually support their AI strategy rather than retrofitting AI into existing systems.
Further Reading
- Zillow bets on content, context, data flywheel as AI differentiator - LinkedIn Pulse
- Zillow's AI Strategy: How the Real Estate Search Leader Built AI Powered Buyer Journeys in 2026 - GetPerspective AI
- Zillow's CTO says AI is reinventing every step of the home buying process - Fortune
- Why Zillow Is Betting on AI to Disrupt Homebuying | Behind the Business - YouTube
