This is the story of a design system that let Google Cloud show every client its own conversational search, in its own brand, before they decided to build it.
People don't think in keywords. They think in needs: a red dress for a beach wedding, under two hundred dollars. Conversational search could meet them halfway, and ask back when it needs to. But in 2023 nobody had seen what that should look like.
Google Cloud's research with The Harris Poll put a number on it. Only about one in ten US shoppers say a retailer's search gives them exactly what they want every time, or a good alternative. Three in four had searched and not found the item.
Google's user research team ran the shopper interviews; I watched the recordings and synthesized them. When search makes you guess the right words, it feels like a test you're failing. The future we were designing had to feel like talking to a helpful clerk instead.
Most of what shaped this came from sitting down with the product managers and UX designers of Fortune 500 retailers, and hearing first-hand how their industries work. Their biggest ask was the same everywhere: more engagement with recommendations, and more cross-selling. The numbers back them. In one Salesforce study, the few visits that clicked a recommendation brought in a quarter of the revenue. McKinsey finds personalization most often lifts revenue ten to fifteen percent. And with online sales passing six trillion dollars a year, a small lift is a lot of money.
That's the case for designing it. Search abandonment costs retailers more than two trillion dollars a year worldwide, and more than half of US shoppers drop their cart after a search that fails them.
And the kit itself had two jobs. For my team, it meant building demos faster, and at scale: this was before vibe coding, so every demo was made in Figma, and switching a variable moved a demo from one industry to the next. For clients, it was inspiration: instead of guessing what to build, they got a boilerplate, a starter kit to run their own experiments.
In the room were CEOs and CTOs, product managers, UX teams and Google Cloud's sales team. What we brought were conversation starters: clickable prototypes that followed a use case and a script written with sales, so stakeholders could feel the story was real and fit their own business. I presented them to prospective clients in New York and remotely, and checked in every week with the teams that were already building.
So instead of designing one demo, we designed a kit. This is one flow, built once, shown in Cymbal, Cymbal Labs and Cymbal Superstore, in light and dark, on desktop and on a phone. Nothing here was redrawn. Color, corners and type all come from the tokens underneath, so every client sees their own future, not Google's.
The complexity lives in the variables. At the bottom, twelve raw hue ramps. Above them, each brand picks its primary ramp and its corners: Superstore is square. On top, every role, from text to surfaces to buttons, maps to a step for light and for dark. Change the brand, and every component follows.
We built it around Cymbal, Google Cloud's demo brand: thirteen sub-brand logos, from a superstore to a bank, and ten industry modes that each swap the primary color. Here the same results page runs through all ten, so a demo could look like the client's world before the first meeting.
Multiply it out and you see the real problem. Five hundred and seventy-two variables in five collections: three brand modes, ten industry modes, light and dark, desktop and mobile, and thirteen logos. Every component has to hold up in every one of those combinations, which is why it had to be a system, not a set of mockups.
The vision was simple: Google Cloud teams walk into a meeting with a working prototype of the client's own search. A future you can touch is a lot easier to say yes to than a slide about it.
The published library was the end of almost two years of building prototypes for different companies. My team and I had UIs scattered everywhere, so we gathered them into one, and decided to make it public, so the whole community could get inspired and build more conversational search.
In October 2025, Google Cloud published the conversational commerce use cases, with the component library on Figma, so client teams could design and build their own conversational search.
Imagine it, make it tangible, pitch it. That's what I did here, and it's what I teach my students too. Thanks for watching.