Zillow, 2026
Rethinking first contact between sellers and agents
Product Design, UX Strategy, Problem Framing, Cross-functional Leadership
I was staffed to build a multi-agent "dating" experience, a way for sellers to talk to a few agents before committing. Three weeks in, the data told a different story. Sellers weren't shopping around. They were struggling to reach the one agent they'd already chosen. So I kept shipping what the team needed, and alongside it I reframed the real problem and wrote the brief that shaped our next bet.
The problem.
Reaching a listing agent meant a bare form: name, address, phone, and no room to say anything. Two-thirds of sellers who tapped "Contact agent" gave up, and the ones we most wanted to keep, the sellers who'd already picked that agent, dropped off the hardest. Of the forms that did come through, 40% weren't real sellers. The same form asked too much of the people we wanted and too little of the ones we didn't.
The reframe.
So I pulled the funnel. 76% of sellers who clicked an agent clicked exactly one, and 90% clicked two or fewer. Line that up with how real estate actually works, most sellers hire the first agent they talk to, and the whole premise flips. Sellers weren't comparing agents. They'd already decided and just couldn't get through, while agents had no way to tell a serious seller from a curious one. First contact was quietly failing both sides. I kept the messaging flow moving and wrote a problem brief that reframed first contact as a two-sided problem, with its own evidence, its own users, and its own bar for success.
The bet
Cut the phone-and-contact friction that scares off real sellers, and make intent an optional signal instead of a gate, so the most serious sellers can raise their hand without filtering everyone else out at the door. I reframed the agent side too, from "too many junk leads" to "no way to tell serious from idle up front," which took the biggest objection off the table before anyone could raise it.
An insight worth naming
A buy-side pilot had already made first contact richer. An optional message step got a lot more people writing custom messages, but lead quality didn't budge. A better message tells you how someone expresses intent, not whether they actually have it. Intent has to be measured directly, on both sides, not guessed from how much someone types.
Designing to be proven wrong
I wrote the brief to be falsifiable, not persuasive: four load-bearing hypotheses, each paired with the thing that would prove it wrong, and an honest flag on the one our consumer data couldn't answer, whether agents would even work a phone-optional lead. I built the research plan around that question, with clear kill, hold, and expand paths.
Impact
Reframed the problem from multi-agent comparison to two-sided first-contact reachability, backed by funnel data, moderated research, and prior-art analysis
Shipped the contact-to-conversation flow across native, mobile web, and desktop, working across four engineering teams
Wrote the standalone problem brief that set the team's next bet, with falsifiable hypotheses and two-sided success criteria that couldn't be gamed by optimizing one side
Set a Figma-to-engineering handoff convention my engineering partner adopted
Co-started a cross-pod AI working group for the seller org that leadership scaled to the full team