We started by reframing the problem. The intake flow wasn't broken — it was working exactly as designed. The design was wrong. Six pages, fourteen minutes, and zero feedback during validation. The form treated patients like database rows, not people about to share medical history with strangers.

Our hypothesis: the form's perceived length was the largest single driver of abandonment. Patients felt the work was endless because each page revealed only what was on it. We bet that progressive disclosure inside a single screen — the user can see the trail, the next step, and the end — would lift completion past 90% even before we touched validation logic.

We instrumented the existing flow first. Every input change, every focus, every blur, every submit — keyed by patient ID and field path. Two weeks of telemetry told us exactly which four fields caused 73% of all drop-off, and which two pages had a median time-on-task above three minutes.

Then we collapsed the six pages into one typed wizard backed by a Postgres schema we modeled around the patient, not the form. Each step is a section that auto-expands as the prior one validates. State lives in URL fragments so users can refresh, share, or come back later without losing their place.

On the backend we built a tRPC layer over a normalized Postgres schema with strict zod validators on both client and server. The in-house team can extend it without us — that was a non-negotiable on day one.

The first prototype was too clever. We had built an animated wizard with motion-driven progress and conditional branching that felt like a product walkthrough. Patients found it disorienting. Three rounds of usability testing later, we cut the animations entirely and replaced them with static section transitions.

The second iteration also taught us something we didn't expect: doctors needed parity with patients. They were filling out the same form during in-person visits. So we shipped a doctor-mode that turns off progressive disclosure and shows everything at once — five fewer clicks for power users.

The final product is quieter than what we started with. That's the point. Forms that look interesting are forms that get abandoned.