Zillow's mission is to make home a reality for more people, connecting over 250 million monthly users to a seamless experience across buying, selling, renting, and financing. But years of strong growth, acquisitions, and strict compliance requirements left data siloed across many tools, at times creating friction for internal efforts at the company. By standardizing on the Databricks Platform with Unity Catalog, Zillow replaced an array of tools with a single governed ecosystem, unlocking self-service insights through Genie and laying a foundation for agentic AI. Lakebase extended that foundation further, giving the AI team a persistent memory layer for conversational experiences. The platform now serves 1,500+ montly active internal users while reducing infrastructure and operational costs by 65% following consolidation.
Challenge: Siloed data and separated systems slowed ambitions
Zillow’s vision for a seamless home shopping experience depended on data moving easily across products, teams, and business lines. In practice, siloed systems, compliance requirements, and sometimes clashing tools made that increasingly difficult.
A connected customer journey created data complexity
Home shoppers expect a single journey. They want to browse a property, book a tour with a real estate agent, and get financing — all without starting over. Zillow's portfolio of products and acquisitions makes that possible, but it also makes the data landscape complicated. Rentals, home shopping, touring, mortgages, and acquired companies each arrived with their own stacks, compliance requirements, and access controls. Across the company, teams were operating many tools independently.
Siloed systems slowed some day-to-day work
A single data integration for a new acquisition could consume months. Meanwhile, the classic separation between operational databases and analytical systems turned routine questions into slow handoffs between teams and tools. And because each silo enforced its own access controls independently, even straightforward requests for cross-domain data required manual coordination between teams, a process that could take weeks.
AI initiatives exposed the limits of the existing architecture
That friction became an obstacle when Zillow began building AI Mode, a conversational experience designed to help buyers and renters explore homes, ask follow-up questions, and take actions like booking tours or connecting with an agent. AI Mode needed low-latency persistent memory for conversations and the ability to evaluate and improve those interactions using the same governed data.
Stitching together a separate database and new pipelines was technically possible, but would have added significant complexity. "If we hadn’t moved to a unified platform, we would have been in a very different position”, said Jaycen Grant, Director of Engineering for Zillow's Data Platform. "When the push toward agentic AI hit, we would have been connecting to five or six separate systems instead of one."
Solution: A unified platform with governance at the core, extended by Lakebase for agentic AI
Zillow solved these challenges by standardizing on Databricks, giving teams a simpler way to govern access, connect systems, and move faster.
Governance became the foundation, not the final checkpoint
Zillow made governance the starting point rather than a final checkpoint, with Unity Catalog replacing the access controls that had kept data locked inside individual business lines. The data platform team implemented consistent tagging, role-based and attribute-based access controls, and structured lab, stage, and production workspaces, enabling teams to move quickly while strengthening compliance and security stances, such as SOX change management requirements, across the board.
Mortgage data, once isolated in a dedicated system, can now be made available between workspaces securely with row-level and column-level masking enforced through Unity Catalog. "We've improved data accessibility and flexibility by removing silos without sacrificing strong security and control," Jaycen noted, "and people don't have to move their data anymore."
A single platform replaced siloed tools and opened the door to self-service
Consolidating into a unified platform gave teams across the company a shared environment instead of isolated toolchains. Data engineers, product engineers, project managers, analysts, and executives all now work within the same platform rather than relying on separate systems managed by separate teams. Where cross-domain data requests once required the platform team to build bridges between systems, teams can now discover datasets, request access through a governed process, and start working without platform team involvement. That shift from ticket-driven workflows to self-service fundamentally changed how the data platform team spends its time, with less focus on maintaining infrastructure and routing requests, and more on building the tools and guardrails that make self-service reliable.
Genie helped teams activate governed data faster
Zillow also extended the value of its governed data through Genie. When Zillow’s SVP of Engineering wanted visibility into internal AI tool usage, Jaycen used Genie Code to identify the relevant tables, generate queries, build a notebook, and publish a dashboard and Genie space in under an hour.
“In the past, creating an executive dashboard like that would have taken days or even weeks. We would have had to track down the right data, pull it together, and refine it before sharing it back. With Genie, I was able to respond within the hour in the same Slack thread,” said Jaycen.
Lakebase brought operational and analytical data together for AI
With the foundational governance in place, Lakebase became an important component of Zillow’s agentic AI stack. Lakebase is a fully managed PostgreSQL database integrated into Databricks and built for agents. For Zillow, it helped bridge the divide between operational and analytical systems that had slowed previous initiatives. Jaycen explains, “We see Lakebase as more than just a database—it’s a new architectural layer that brings together transactional systems and AI-driven applications. Because it’s Postgres, teams can adopt it quickly, but its real value is in what it enables beyond traditional databases.”
The AI team deployed Lakebase as the agent memory layer for AI Mode, providing a high-concurrency, low-latency store for conversation history and context. Because Lakebase is integrated with the lakehouse, the agent can retrieve relevant history, then run evaluations, tracing and analysis on the same data without moving it to a separate system.
Because Lakebase inherits Unity Catalog’s compliance framework, the team was able to access regulated datasets, including financial and credit information, under the appropriate governed controls and without additional months of security review. “Without Lakebase, we would have spent months going through security and compliance reviews before we could use those datasets. With Lakebase, they were available right away,” Jaycen explained.
Results: A governed foundation for data at scale
Consolidating onto a single governed platform expanded access across data engineering, application development, project management, analytics, and executive teams, while freeing the data platform team to focus on higher-value work. Operational overhead dropped by more than 65%, and the platform grew to over 1,500+ active monthly users. Even with 450% user growth, platform support tickets per active user fell 44% as self-service replaced ticket-driven workflows. Genie accelerated that self-service shift further, compressing tasks like building an executive dashboard from days or weeks to under an hour.
Lakebase gave Zillow a practical path to production-grade, compliant agentic AI. A single Postgres-backed memory store now supports both real-time customer experiences and offline evaluation, while giving AI Mode access to sensitive datasets that previously would have required months of review. Future agentic applications can now follow the same governed pattern instead of starting from scratch. That foundation is already visible in Zillow AI Mode, which launched publicly in March 2026 as a conversational experience designed to guide users through more of their housing journey.
Key results include:
Operational overhead decreased by more than 65%
A single memory store that supports both real-time experiences and offline evaluation
Sensitive datasets are now accessible through a governed framework
Platform user support tickets per active user fell 44%
Zillow plans to expand Genie into finance, HR, and sales, and apply Lakebase to additional agentic experiences beyond AI Mode. With less time spent on infrastructure maintenance, the data platform team can focus on embedding compliance, cost governance, and data quality directly into the tools people already use. “Databricks takes care of the underlying infrastructure so our team can focus on the work that matters most,” Jaycen said. “It feels like we have an extra development team helping us move faster, and new capabilities keep opening up even more opportunities to expand.”