Zillow Builds AI Architecture to Track Customer Journeys

Key points

• Zillow touches 80% of U.S. real estate transactions annually with its technology.
• Zillow's AI strategy focuses on context management over data collection.
• Glean's integration helps Zillow reduce AI costs and improve efficiency.

Zillow, the well-known real estate technology company, is tackling a unique challenge: keeping track of customer journeys over long periods. This is no small feat, given how people often move from browsing on phones to speaking with loan officers or real estate agents. At the VB Transform 2026 event, reported by VentureBeat, Zillow's SVP of Engineering Toby Roberts and Glean's CEO Arvind Jain shared insights on building an AI system that preserves context throughout this process.

Zillow manages a significant share of the U.S. real estate market, interacting with 80% of transactions each year. They've used artificial intelligence (AI) for years, even before tools like ChatGPT became common. Roberts explained that while many AI projects start with data, Zillow found that maintaining customer context was the real challenge. This context layer ensures that customers are supported no matter how they interact with Zillow.

Instead of relying on a single AI model, Zillow opted to create a system tailored to specific tasks. They decided to keep control of this context-tracking layer internally, rather than depending on any external chatbot. This choice allows them to handle the complexity of real estate transactions more effectively.

Glean plays a vital role in Zillow's strategy by centralizing the integration work. Jain highlighted how Glean's system helps by routing tasks to smaller, cheaper models, cutting down on the resources needed for AI processing. This not only reduces costs but also speeds up the AI's performance.

What does this mean for enterprises?

Roberts emphasized that for enterprises looking to develop AI systems, it's crucial to have a solid measurement baseline in place before launching AI initiatives. This approach allowed Zillow to credit a 40% increase in its code output to AI, thanks to pre-existing metrics rather than post-hoc calculations.

Jain pointed out that centralizing context can save enterprises from hidden costs associated with repeated integration work across different departments. Additionally, enterprises should be cautious about assuming that permission systems alone will manage sensitive data. Zillow, for instance, added strict compliance checks to safeguard its customer data.

By focusing on context, not just capability, Zillow's AI strategy serves as a model for other businesses. Centralizing context helps reduce AI expenses, offering a practical lesson on how to integrate AI effectively within a company.