Nobody left because the form was too long.
That's what I assumed, at first. Our onboarding intake had fifteen fields — insurance, prior diagnoses, presenting concern — and completion was under 30%. The obvious fix was to shorten it.
Then I watched a session recording of a 17-year-old typing "i dont know if im depressed or just tired" into a field labeled Primary Concern, deleting it twice, and closing the tab. She hadn't left because the form was long. She'd left because we'd asked her to diagnose herself before we'd given her any reason to trust us with the answer.
Every recording stalled on the same nine words.
I ran five test sessions each with three personas built from real intake data — Maya (17, academic burnout), Raj (15, avoids strangers and video calls), and Anjali (22, first time seeking care). I wasn't looking for a pattern going in, but one showed up in four of the five recordings.
It wasn't a field, a button, or a load time. It was the exact moment the form asked "What brings you here today?" — before anything on screen had told the user their answer was safe, private, or judged. Raj, specifically, never made it past that field into the video-call prompt that followed it.
Left: the full journey map. Right: the same map annotated at the one point every stalled session shared.
We moved the hardest question to last.
The fix wasn't to soften the wording of that one field — I'd tried that, and it didn't move completion. The fix was to stop asking it first. In the new flow, someone can read peer stories, browse a mood check-in, and sit in a community space before the product asks them anything that sounds like a diagnosis.
By the time the app does ask what's going on, it's phrased as an invitation instead of an intake field: "Let it out" instead of Primary Concern, "Start when you're ready" instead of Book Intake Appointment. Same information, asked at a point where trust already exists.
The community had to feel like a group chat, not a forum.
Raj's interviews kept surfacing the same tension: he wanted to talk to someone about how he was feeling, but a live video call with a stranger was the one thing he wouldn't do. Existing peer options were either unmoderated, forum-style servers where a bad post could sit for hours, or nothing at all.
We built small, topic-based rooms instead — anonymous by default, text-first, moderated on a schedule tight enough that nothing concerning sits unanswered overnight. No profile photos, no follower counts. Just a room named for what someone's dealing with, and other people already in it.
A grid of therapist headshots is a decision no one wants to make first.
The original therapist directory was a searchable grid — filter by specialty, insurance, gender, price. In testing, people spent ten to fifteen minutes comparing credentials without booking anyone. More options had made the decision harder, not easier, for someone who was already running low on the capacity to make decisions.
So we removed the grid for first-time users. Based on what they'd already shared during onboarding, the app now surfaces one considered match at a time, with the option to see another — never a full list up front. It reads less like a marketplace and more like being introduced to someone.
The clearest sign the rebuild worked wasn't a metric — it was that support messages asking "is this thing actually private?" nearly disappeared. Once people stopped being asked to diagnose themselves in field four, they stopped needing to ask us if it was safe to keep going.