How mixed-methods research turned an ambiguous authentication transition into a clearer experience — and a measurable lift in completed payments.
The product question
Users were leaving a merchant experience and moving into bank authentication. The screen between those moments needed to explain the transition quickly, reduce uncertainty and preserve momentum.
What must people understand before they feel ready to continue?
Method followed the decision
I separated exploration, validation and behavioral measurement so each stage could answer a different question — without overstating what any single method proved.
Qualitative exploration
Fourteen moderated interviews across the United States, United Kingdom, Germany, France and Sweden revealed how people interpreted the handoff — and where meaning broke down.
Stated-preference validation
A structured validation with 102 participants compared wording in the US and UK, showing a clear preference for language that named the verification action.
Production A/B test
The product team tested a shorter, clearer communication hierarchy in production. The experiment measured the resulting experience, not the isolated effect of a single word.
The mental-model gap
People needed the screen to answer four questions before the next step felt legitimate.
What am I doing?
Why am I doing it?
What am I approving?
Why should I trust this?
Language as product behavior
The validation did not choose copy by taste. It tested whether the words matched the task people believed they were about to perform.
Names the security action and gives the handoff a reason.
Signals navigation, but leaves the purpose of the transition unclear.
Research as a decision system
The production experiment
The variant combined a new headline with less content, a focused explanation of authentication and explicit reassurance about credential privacy.
Scope of the claim: the A/B test evaluated the complete communication hierarchy. It did not isolate the causal effect of one word.
Behavioral result
The result was internally confirmed as statistically significant.
AI with evidence boundaries
Canonical records preserve source, market, sample, quotations and metrics.
Gemini and Claude workflows support synthesis, documentation and retrieval from approved evidence.
I frame the question, challenge the output and keep claims inside what the evidence can support.
The same research logic applies wherever people cross a boundary they cannot fully see: wallets, P2P transfers, bank linking and authentication flows.