One landing page that adapts to who the seller is
Wix was losing dropshipping sellers to churn and had never reached one big group at all: site owners with no online store. My answer was a "Find Products to Sell" page that reshapes its entire purpose based on where a seller is in their journey — designed, built, and handed to engineering as a working AI-assisted prototype.
A single page can't serve four very different sellers
Wix's "Find Products to Sell" page treated every visitor the same — but the people landing on it were anything but. Some had no online store at all. Some had just opened one. Some were running mature catalogs. A page optimized for one of them actively underserved the others.
The redesign brief was to lift engagement and return visits. My read was that the lever wasn't a prettier page — it was a page that knew who it was talking to and changed its entire purpose accordingly.
Four states, four jobs
Rather than one layout with conditional bits, I designed four genuinely different experiences — each with its own hero, content, and goal, matched to the seller's maturity and intent.
Reached via contextual prompts where intent already exists — adding products to a booking service, or swag for an event. AI reads their site content and previews their logo on the top 5 recommended POD products.
A fixed setup-guidance area, echoing the Wix setup dashboard, focused on getting payments, shipping, and the right number and type of products in place. No catalog optimization yet — there isn't enough data to be useful.
Weekly, data-driven suggestions: restock low inventory with similar items, add complements to top performers, surface industry-trending products, and — at sufficient volume — recommend branching into dropshipping, POD, or wholesale.
States 2 and 3 further adapt to what the seller has already added: have POD? suggest dropshipping, and vice versa. Tried nothing yet? Lead with POD — the easiest on-ramp into selling products.
Wireframes above are simplified representations of the four states, created for this case study. The full-length “before” capture in Context is the live production page (2026).
From idea to a working prototype — solo
The part I'm proudest of isn't a screen — it's how far I could carry this alone. I used AI not as a novelty but as the actual production pipeline, moving from concept to a running, API-connected prototype without a separate engineer for the foundation.
A detail I care about: once the work moved into code, code became the source of truth. I stopped going back to Figma. When Wix landed a new POD partnership (Blanka) mid-build, I implemented it directly in Claude Code and Cursor, and I wired the prototype to Wix's product APIs so it pulled real products — not placeholder mockups.
One of the AI-generated passes
Below is a raw Figma Make exploration of the four-state concept — the kind of quick generated pass I use to pressure-test an architecture before committing to a direction, recreated for this case study. It is deliberately not the shipped design: the sketchbook, not the ship. Switch states to feel how the concept adapts. Best explored on desktop.
What this project demonstrates
- Systems thinking over screens — designing intent-aware states, not a single page, to serve four very different sellers.
- Product ownership — KPIs, scope, research, and backlog owned end to end in a PM capacity.
- Real AI-to-production fluency — moving from concept to an MCP- and API-connected prototype, and handing engineering a running PR instead of a static file.
- Honest collaboration — building on my manager's AI-recommendations direction, and clear about where my build handed off to engineering.
The project was handed off as a PR foundation; my role ended before it reached production.
