Designing a physical product category from zero

In late 2022 Fullscript decided to sell something it had never made: its own product, personalized strips of daily supplement pouches filled and shipped under the Fullscript name. A fulfillment partner handled the physical side, but the agreement required the whole experience to read as native Fullscript — no co-branding to lean on, no partner documentation to link out to. The design surface was the product.

I wrote the relevance boards that defined who this was for — virtual prescribers, where 42% of prescriptions carry exact dosages — and mapped the four core flows before the first build sprint. When the product pages needed a spec, I wrote the requirements doc myself; the release notes credit it by name.

Early explorations put packs in the catalog, and we cut that to reduce scope — custom packs exist per practitioner, a concept the search infrastructure couldn't hold. Packs shipped in their own tab instead, with a direct action that opens a draft plan. It was the buildable option and the invisible one: nobody could find packs, and most of my later work went into buying back what that cut cost us.

A first-party product with an invisible fulfillment partner needs a trust anchor, and leading with a brand nobody knew meant explaining a brand nobody knew. So the person patients already trust does the explaining: "Personalized for you by Dr. {Name}" replaced the brand line, backed by an Only on Fullscript badge and a product page generated per patient.

The fulfillment facility could print 27 characters, legal owned what a practitioner could type as a pack name, and inventory was fragile enough that we held back a whole pilot wave for it. Personalization kept meeting physical limits — the print constraint cut practitioner names from the booklet one day before launch.

Three people, an unprecedented category, and no live users to test against — formal studies would have burned weeks testing guesses. We built the first version on judgment and measured everything once it was live. A dress rehearsal placed real orders through the pipeline in May 2023, then 50 practitioners, then 400 by mid-June, with every release killswitch-able in case inventory ran low.

Two weeks after the August launch the numbers were blunt: a handful of practitioners had explored the builder, and almost none had prescribed. Real usability truth arrived only after launch, and it said first use was the problem. Every design move that followed — a guided tour, an awareness banner, templates — targeted that first try.

Sales hand-picked 50 ideal-profile practitioners for wave one and got a single adopter out of them, while 350 randomly chosen accounts produced eighteen. Every holdout asked for the same thing — let me customize it — and that became the entire second release. When leadership paused templates, our memo reversed it within a week: the design was already done and kept component-fresh, which made saying yes cheap.

The invisibility problem outlived the product. Practitioners have since built hundreds of thousands of curated collections on the platform, and the platform's own numbers put all of them together at under 2% of direct order volume — because the overwhelming majority of patients arrive on a direct link to their plan and never pass a storefront at all. Same cut, same result, at a scale I couldn't have imagined.

A physical category is designed everywhere except the screen. The strongest re-order prompt we shipped was an empty pouch, and the deliverables that made this product real never appeared in a Figma file.