Key findings
- Document rejection was partly a configuration problem.
- Requests for personal identifiers and tax residency information were not adequately explained.
- The risk questions felt like an exam to some participants.
- Leaving to find a document also meant losing progress.
- ID checksneeded, kept
- Photo captureglare, rejections
- Leave to find IDprogress lost
A shorter sign-up.
The checks weren’t the problem. Photo capture and lost progress were.
At a glance
The client was an investing app extending its service into a new market. People were abandoning account setup, and an external identity-verification supplier handled document and selfie checks. We investigated the sign-up with the product team, compliance, fraud specialists and the supplier.
Our role combined research in the circumstances where people signed up with analysis of the system's failure records. The objective was to distinguish avoidable difficulty from checks that needed to remain. A shorter sign-up was only useful if it still performed the job the service required.
The brief
The client saw weaker sign-up completion in the new market and asked us to make onboarding shorter. That was an understandable request, but it combined several different problems. A rejected document, an unexplained tax question and a person deciding an investment product was unsuitable could all appear as abandonment in a funnel.
We therefore agreed to examine what each step was trying to accomplish before deciding whether to remove it. Some steps needed better explanation. Others needed a supplier configuration change. Some needed to help a person make an informed decision even when that decision was not to continue.
The constraints that shaped the work
The compliance team owned the requirements for identity and product-risk questions. We worked with them at the start rather than taking a finished list of conversion ideas to them later. That made it possible to distinguish requirements from conventions the interface had accumulated.
The appropriateness questions were a particularly important boundary. Helping someone understand a question was different from helping them choose the answer that would let them proceed. We agreed that comprehension would be the research measure, not a higher pass rate.
Fraud specialists also needed to be part of the evaluation. An easier capture flow would not be an improvement if it admitted unsuitable evidence. Manual review and fraud indicators remained part of the assessment alongside completion. The third-party supplier meant that some changes required negotiation and configuration, not just new interface copy.
How we investigated the sign-up
We broke the funnel down into individual steps and included the identity supplier's pass and failure codes. This helped us avoid treating every unsuccessful attempt as a user choice to leave. A technical rejection could look like disinterest unless the two records were considered together.
Moderated sessions took place on people's own phones in their home environments. Document capture and selfie checks depend on the device, lighting and documents someone has available. A clean office demonstration could miss the conditions causing difficulty during an evening sign-up.
We reviewed support tickets to find repeated attempts, confusion and workarounds that were not apparent from a completion rate. Working sessions with compliance, fraud and the supplier then connected the findings to the people who could change the relevant part of the sign-up.
What we found
Document rejection was partly a configuration problem. The accepted-document settings and capture behaviour did not fit every document people reasonably tried to use. Glare and low-light selfie capture added further difficulty. Asking people to try again without explaining what to change made a failed attempt feel arbitrary.
Requests for personal identifiers and tax residency information were not adequately explained. People could interpret a legitimate information request as a sign that something was wrong with the app. The interface assumed familiarity with the purpose of a question that the person completing it did not share.
The risk questions felt like an exam to some participants. They searched for an answer that would let them through rather than using the questions to understand the product. A higher completion rate would have concealed that failure of comprehension.
Leaving to find a document also meant losing progress. The sign-up treated a practical interruption as a restart. That made the absence of an accepted document at the first attempt more costly than it needed to be.
What we tested and recommended
We tested a preparation checklist that explained what documents to have ready and what capture conditions would help. This moved useful guidance before the point of failure, without requiring people to learn the entire process in advance.
We worked with the supplier on accepted-document settings, capture support and glare detection. This was a different intervention from rewriting the error message: the research had identified a problem in the service configuration as well as its presentation.
For personal-information questions, we tested short explanations of why the information was being requested. We also evaluated save-and-resume so that someone could gather what they needed and return. The aim was to make an interruption recoverable, rather than to pretend that everybody would have every document at hand.
The risk questions were revised and retested for comprehension with compliance involvement. We did not use a higher pass rate as proof of better design. Throughout the work, changes to regulated screens followed the team's sign-off process and were considered alongside the fraud indicators.
The principle
Required friction and accidental friction need different responses. Remove the difficulty that serves no purpose; explain the checks that do. A conversion improvement is not meaningful if it leaves people less able to understand what they are agreeing to.