July 8, 2026 · 3 min read

The onboarding fix was letting people skip it

In 2023 I designed the onboarding flow for Mentor Buddies, a mobile app by Design Buddies — one of the largest design communities — that matched members with expert mentors for short private chats, in exchange for a small donation to the mentor's chosen charity. The app had already been through a beta test. The completion rate for onboarding was 1.9%.

Not retention. Not conversion to paid. Onboarding. Ninety-eight of every hundred people who wanted mentorship gave up before seeing a single mentor.

Finding where they left

The analytics gave us the where before the interviews gave us the why. Three screens accounted for most of the bleeding: the welcome screen lost 35% of arrivals, profile creation lost 66% of those remaining, and the mentor list lost 66% again.

To understand the why, I interviewed eight community members and ran affinity mapping on their feedback in FigJam — every comment tagged to the screen it referred to, then clustered into themes. Four pain points survived the clustering:

  1. Users weren't invested enough to hand over personal information at account creation.
  2. The welcome screen was a wall of text; people couldn't tell how the app worked.
  3. The mentor list was overwhelming.
  4. Matching felt neither accurate nor personal.

The first one is the one that matters, because it isn't a UI problem. It's a sequencing problem. The app was demanding commitment — a filled-out profile — before it had demonstrated any value. Users were being asked to invest in a product they hadn't seen yet.

The uncomfortable fix

The instinctive response to a failing profile form is to improve the form: fewer fields, better copy, progress indicators. We did some of that. But the change that actually addressed the pain point was structural: an alternative path that skips profile creation entirely and drops the user straight into the mentor list.

This was not an easy sell to stakeholders — a profile is data, and data feels like the business model. The argument that won was the same data pointed the other way: a 66% drop-off on that one screen meant the profile requirement was destroying two-thirds of the funnel to protect information we could collect later, from users who by then had a reason to give it. Once someone has browsed mentors and wants to book a session, the profile stops being a toll booth and becomes a natural step.

The other fixes followed the same principle — show value before asking for effort. The text-heavy welcome screens became illustrations that explained the mechanic (expert chat, 15 minutes, donation to charity) at a glance. The mentor list got reviews, self-intros, and online-status indicators, so the value being promised was visible and specific rather than abstract.

Testing and the number

I sketched the new flow on paper — no more than two major interaction areas per screen — moved to a low-fidelity Figma prototype, and ran think-aloud usability tests with five participants whose task was to get from install to chatting with a mentor. The structure held; the flaws it surfaced were fixable details, not direction.

After development, the team ran a private beta with 125 community members. Onboarding completion went from 1.9% to 30.4% — a 16× improvement, with the previously fatal drop-off points resolved.

What I took from it

Drop-off data tells you where; only users tell you why. The analytics pointed at three screens. The interviews explained that one of them wasn't a screen problem at all.

The best onboarding step is often the one you make optional. Every mandatory step is a bet that its value to you exceeds its cost in abandoned users. At a 66% drop-off, that bet was losing badly — and moving the step later in the journey converted it from a barrier into a formality.

Designs survive contact with engineers only if engineers are in the room. With no PM on the team, I worked with the developers daily, designed edge cases on their request, and kept everything on an 8pt grid so consistency didn't depend on a design system we didn't have. The 30.4% belongs to that collaboration as much as to any wireframe.

Related Writing