Analytics showed a clean funnel with one obvious problem: a third of people who started signing up never got past the third screen. Everyone had an opinion about why. The data, once we looked properly, was less interesting than any of the theories.
What the screen actually asked for
The third screen asked new users to connect a bank account before they had done anything else in the product. It was a reasonable request from the business's point of view — the integration mattered a lot to the value proposition — and a large ask from a stranger who had been using the product for ninety seconds.
What we tested instead of redesigning
Rather than restyling the screen, we moved the request. Bank connection became optional at signup and required only at the point a user tried to do something that actually needed it — several steps and, for most users, several minutes later, once they had already seen enough value to trust the request.
- Signup completion was measured before and after, on the same traffic mix, over a comparable period.
- The connection rate itself was tracked separately, further downstream, to check we had not just relabelled the drop-off rather than removed it.
- Support tickets about the bank step were tracked too, since a confusing later prompt could have created a different problem.
The result
Signup completion rose sharply, and — the part that mattered to the business — the eventual bank connection rate, measured across the whole user lifecycle rather than just at signup, held steady. Moving the ask did not reduce how many people eventually connected an account; it just stopped losing people before they had any reason to trust the request.
The right point to ask for something is not determined by when the business wants to know. It is determined by when the user has enough reason to say yes.
Why this was not really a design problem
No amount of copywriting or visual polish on the original screen would have fixed this, because the problem was sequencing, not presentation. A well-designed version of the wrong question at the wrong time is still the wrong question at the wrong time.
What we now check earlier
We now map every onboarding step against what the user has actually experienced by that point, before a single screen is designed, specifically looking for requests that outrun the trust the product has earned so far.