Onboarding that stops losing people

The problem

Opening a bank account should take minutes. In practice, 38% of people who started the application never finished it. Support tickets pointed to the same three moments: identity verification, address entry, and the final confirmation screen.

The team had already shipped two rounds of copy changes. Completion barely moved. That was the signal that the problem was structural, not verbal.

What research changed

We ran eight moderated interviews with people who had abandoned the flow in the previous 30 days. I synthesised the sessions into a journey map and a short list of moments where confidence dropped.

Three patterns came back repeatedly:

• People could not tell whether the app was working or waiting for them.
• The address step asked for information the app already had.
• The final screen looked like a receipt, so people assumed the application had failed.

“I wasn’t sure whether the app was checking my documents or waiting for me to do something.” — Participant 4

The decisions that mattered

We made waiting visible. Identity verification stayed with the third-party SDK, but we added a persistent status line with plain-language stages. People still waited, but they stopped guessing.

We removed the address step for existing customers. The app already had the data. Asking again was a trust problem disguised as a form problem.

We redesigned the confirmation screen around the next action. Instead of a receipt, the screen now says what happened, what is being checked, and when the account will be ready.

We cut the progress indicator. It implied a fixed number of steps, which was never true when verification took longer. A status model was more honest than a progress bar.

What shipped

The release covered the mobile application flow in Poland. The team shipped the new interaction model behind a feature flag, compared it with the previous flow for four weeks, and rolled it out to all users.

What I’d do differently

I would bring engineering into the status-model exploration earlier. We spent a week designing states that the SDK could not expose, and had to simplify late. The next time I would prototype the state machine with the engineer who owns the integration before drawing the screens.

I would also test the confirmation screen with people who had never used the app, not only with people who had abandoned it. The first group read the screen very differently.