The Setup
FinFlow, a mobile banking startup, redesigned their onboarding flow to modernise their UX. The result was unexpected: completion rates dropped from 78% to 61% — a 17-point decline.
The internal debate was familiar. Product blamed the compliance team for adding KYC friction. Engineering blamed the UX redesign. The CEO wanted data by the next board meeting.
They had two things: 8 user interview transcripts and funnel analytics for 40 users who went through the new onboarding. This is what FableSense AI found.
The Data
What we had:
- 8 user interviews (Marcus, Priya, David, and 5 others)
onboarding_funnel.csv: 40 users, tracking ID scan attempts, completion time, NPS at day 7, 30-day retention, monthly active days, support tickets
| Metric | Value |
|---|---|
| Overall completion rate | 61% (down from 78%) |
| Average completion time | 31 minutes |
| Avg NPS — 1 scan attempt | 76 |
| Avg NPS — 4+ scan attempts | 24 |
The Investigation
Qualitative Coding
We built a code framework around the themes emerging from the interviews:
- ID Verification Issues — camera problems, multiple attempts, giving up
- Security Concerns — hesitation about linking bank accounts, privacy worries
- Time Pressure — attempting onboarding in the wrong context, getting interrupted
- Positive Experience — quick completion, would recommend
The first interview — Marcus, a small business owner — made the pattern clear immediately.
"The camera kept saying 'image unclear' even though I was in good lighting. I tried like six times."
"I gave up that day. Came back two days later when I was less annoyed."
Marcus eventually completed onboarding. But it took 47 minutes and left a mark. His day-7 NPS: 31.
Priya's story revealed a different but related failure mode:
"I had to scan my passport. I don't carry my passport around."
"I had to wait until I got home. By then I kind of forgot."
She never came back. The onboarding assumed users had their documents on hand, in good lighting, with time to spare.
David added a third dimension — the UX itself was unclear:
"I didn't understand why I needed to take a photo of myself. It felt invasive."
"The 'move your head slowly' instruction — I didn't understand what it wanted."
Contrast all of this with Nina (U009):
"Super easy for me. Did it in bed one night. The face scan worked first try. I was done in maybe 7 minutes."
Same app. Same step. Completely different experience.
The Joint Display: Where It Came Together
Individual Level: Marcus vs. Nina
Using FableSense AI's Case-Level Analysis joint display, we pulled each user's qualitative profile alongside their quantitative metrics — side by side.
Marcus (U002):
- Themes: Multiple attempts needed, Gave up temporarily
- ID scan attempts: 6
- Completion time: 47 minutes
- Day-7 NPS: 31
- Monthly active days: 12
- Support tickets: 1
Nina (U009):
- Themes: Quick completion
- ID scan attempts: 1
- Completion time: 9 minutes
- Day-7 NPS: 78
- Monthly active days: 26
- Support tickets: 0
Same product. Same step. 5× difference in scan attempts. 2.5× difference in NPS. 2× difference in monthly engagement.
Population Level: The Integration Matrix
The Integration Matrix confirmed this wasn't anecdotal. Mapping every theme against every metric revealed:
| Theme | onboarding_completed | day7_nps | day30_retention |
|---|---|---|---|
| ID Verification Issues | Strong negative | Strong negative | Strong negative |
| Gave up temporarily | Strong negative | r = -0.788 | Strong negative |
| Quick completion | Strong positive | Moderate positive | Moderate positive |
The statistical view surfaced the headline finding:
"Gave up temporarily" × id_scan_attempts: r = 0.934
A 0.93 correlation. Near-perfect. The more scan attempts a user needed, the more likely they gave up. And giving up correlated with NPS cratering.
"Negative First Impression" × support_tickets: r = 1.398
A bad onboarding experience didn't just lower satisfaction. It drove support load for months afterward.
The Recommendation
Three specific, prioritised fixes came out of this analysis:
1. Fix the ID scan UI (highest impact) Add ambient light detection before prompting for the ID scan. Show progress ("attempt 2 of 3") so users don't feel like the app is broken. Provide an alternative path for users who need to return later.
2. Add contextual prompts before the ID step Warn users they'll need a government-issued ID and a private space before they start onboarding — not when they're already frustrated on step 4.
3. Segment the follow-up Users who abandoned and came back within 48 hours (like Marcus) are recoverable. Trigger a re-engagement nudge with clearer instructions before they give up for good.
The Outcome
This analysis took less than a day using FableSense AI. The qualitative explained why — friction at the ID step destroyed the early experience and damaged long-term engagement. The quantitative proved how much — users with 4+ scan attempts scored 52 NPS points lower and generated 3× more support tickets.
The team had their answer before the next board meeting.