Process
The same six steps, every product.
Whether the project is two weeks (Twine’s retirement visualizer) or two years (Clip’s payment platform), the underlying loop is the same. AI compresses synthesis, documentation, and prototyping — it doesn’t replace the rigor.
- 01
Research
Stakeholder interviews, on-site observation, journey mapping. Get out of the office and into the merchant’s shop, the member’s club, the user’s couch.
For Clip I flew to Mexico City and ran merchant + consumer workshops. For Sam’s I sat with a researcher in a Concord club. For Twine we audited 15 retirement calculators on the market. AI helps synthesize notes and transcripts afterward — the discipline of seeing the user in their context doesn’t change.
- 02
Synthesis
Cluster findings, articulate insights, frame the problem. Translate research into something a cross-functional team can act on.
Findings on a wall, posted notes, clusters, headlines. The point is to leave with a thesis the cross-functional team can argue with. At Sam’s the synthesis surfaced three buckets: member engagement, shopping, and "most loved membership." AI accelerates the clustering; the judgment call is still human.
- 03
Ideate
Sketch first, never start in the computer. Diverge wide, share the wall, get feedback early before committing pixels.
Notebook first. Then large sheets on the wall and a critique. For Clip I jumped straight to HiFi to compress the time-to-engineering hand-off; for Twine I diverged twice before converging on a quiz-first format. AI is useful for variation and copy drafts — convergence still happens in the room.
- 04
Prototype
Lo-fi to hi-fi as the story sharpens. Printable, clickable, coded — whatever the question demands.
Lo-fi paper, clickable InVision, coded Framer/Principle, or in Twine’s case the built-in Figma prototyping when the timeline only allowed two weeks. AI-assisted tooling (including this portfolio site) shortens the path to something testable. The right fidelity is still the one that answers the next question.
- 05
Validate
Test with real users in their context. Iterate on what we learn. Repeat until the experience clears.
For Clip we used monthly pop-up mercados in Mexico City. For Sam’s our researcher ran our prototype in the club twice. For Twine we tested side-by-side against NerdWallet. Real users, real context, real friction — no substitute.
- 06
Metrics
Tie the work back to business outcomes. Engagement, conversion, GMV, NPS. Design that doesn’t move the needle isn’t finished.
Frequently Ordered Items at Sam’s was a $37.4M annualized GMV lift. Clip processed $1B+ pesos. Twine had 3x Apple App of the Day and a 4.6 rating. Design that doesn’t move the business doesn’t survive.
AI in the practice
Faster, not shortcut.
AI is integrated throughout the loop below — not a separate phase. These are the places it actually earns its keep.
Compressing interview notes, workshop output, and competitive audits into themes — then stress-testing them with the team.
Accelerating first drafts of flows, copy, and specs. The thinking stays human; AI handles the repetitive lift.
Spinning up functional prototypes faster — this site included — so validation happens sooner.
Knowing where AI helps, where it hallucinates, and when to ignore it entirely.
See it in action. Browse the case studies — including AI work at Workday — or jump into the Clip Terminal write-up for the long version.