Products that work. Proven with numbers.
Across consumer marketplaces, fintech and B2B platforms used by millions. End-to-end — from ambiguous problem to shipped, validated product — with conversion, task completion and retention as the closing argument.
Let's talk shopI trust process more than inspiration. The strongest work I've shipped started in a spreadsheet or a customer support ticket — not a moodboard. The ideas I've loved most have, more than once, been killed by the data, and that's been good for the products. Forming a hypothesis, testing it, being willing to be wrong in public — that's the most honest way I know to make something that actually works. Inspiration shows up plenty. It just doesn't get a pass on the rest of the work.
How I work
Data.
Every claim I make about a design closes with a number. Hypotheses are the start of a project, not a postscript. Measure, hypothesise, prototype, test, learn, ship — and the willingness to be wrong in public is what makes the next thing better.
Business.
Design that doesn't pay for itself doesn't ship. I sit between business and engineering, translate in both directions, and treat margins, payback periods and operational constraints as part of the brief — not someone else's problem.
People.
The user is the only stakeholder who can't be in the room. I keep their interests visible against pressure to ship faster, simpler, or with the friction tilted to favour the operator. In regulated work that means compliance is a feature, not a tax. In commercial work it means a real choice, not a soft sell.
AI.
A tool to be understood, not a posture to be adopted. I worked on conversational design before it was called AI — automotive work at IBMiX, when the problems were turn-taking, fallback states and intent disambiguation. The tech changed; those problems didn't. I bring that frame to LLM-era work: clear-eyed about what the tools are good at, and honest about what they aren't.