Finding the
real bottleneck
Top-of-funnel was improving, yet activation wasn't following. I pushed the team past aggregate conversion into weekly cohorts and found the KYC completion step quietly bleeding, hidden inside averages. The fix wasn't a feature; it was an operating model.
01Situation
A funnel that looked healthy from the top
Self-serve onboarding at a regulated B2B fintech: signup → company created → KYC submitted → KYC completed → wallet loaded. After Q1 product changes, the early funnel was genuinely improving: Lead → Company Created → KYC Submitted all trended up. On the surface, a success story.
02Problem
Activation wasn't following
Downstream, KYC completion and wallet load stayed stubbornly flat. The funnel was filling faster but not finishing better. Aggregate conversion numbers hid where and when the leak was happening, so every team argued from anecdotes: was it traffic quality? UX confusion? Policy? Analyst backlog?
03Goal
Move activation, not just signups
Turn "conversion is down" into a precise, shared diagnosis (which stage, which cohorts, which cause), then focus six teams on the single highest-leverage constraint instead of six separate guesses.
04My role
Product Lead. I drove the shift from aggregate to cohort analysis, built the shared funnel definitions, and aligned Product, Data, KYC/CDD, Ops, Marketing and Engineering on one diagnosis. I didn't own every fix. I made the system visible so the right teams could own theirs.
05Team
Cross-functional by necessity: Product & Eng (flows, instrumentation), Data (cohort models), KYC/CDD (policy & checks), Ops (analyst capacity), Marketing (traffic quality). Onboarding wasn't one team's system. That was the point.
06Diagnosis
Cohorts, not averages
I pushed the team to cut the funnel by weekly signup cohorts rather than aggregate conversion. That's when the leak showed: in March, KYC submitted → completed dropped materially for specific cohorts, from 50.7% down to 44.2% and 43.5% in affected weeks.
The cause wasn't one thing. More cases were entering feedback loops and blocked states, and customers who entered a loop often didn't return in the same week, which dragged down completion and everything downstream.
07Decision making
Fix the operating model, not a screen
The tempting move was another UI iteration. The data said otherwise: the constraint sat between teams: definitions, feedback-loop policy, operational latency, traffic mix. So the decision was to treat onboarding as a system: align every team on the same funnel definitions and stage ownership, then prioritise interventions by leverage, not by who shouted loudest.
08Solution
One system, four interventions
09Trade-offs
What we deliberately didn't do
No blanket friction cuts. Removing checks would lift completion and raise risk, and the constraint was surprises, not scrutiny. No single-team hero fix. A UI-only optimisation was cheaper and faster, but the data showed it would miss policy, ops and traffic causes. Slower, system-level alignment was the honest path.
10Impact
From argument to aim
The org stopped debating whether conversion was down and started working a named constraint: KYC submitted → completed, specific cohorts, loop-driven. Interventions were prioritised on evidence; the early-funnel gains (see the signup experiments) could finally compound into activation instead of stalling mid-funnel.
11Learning
Regulated onboarding is a system, not a screen
If you optimise only the UI, you miss policy, ops, communication, data quality and traffic mix. The Product Lead's job is to make the whole system visible, then focus teams on the highest-leverage constraint.
"We moved from 'conversion is down' to a system-level diagnosis."