Volta — Funnel Analysis
Situation
The first project in the discover loop: find where the fictional neobank «Volta» loses users in onboarding. Data is synthetic, seeded generator → reproducible CSV.
Task
I owned the reproducibility of the dataset and the step-by-step measurement of loss, rather than one aggregate estimate.
Actions
I broke the funnel down step by step so the loss is visible at every transition, not just at the end of the path.
- Data assembly. Built onboarding events from the seeded generator and shaped them into a ‘user × step × status’ table, so the steps are comparable to one another.
- Step metrics. For every transition I computed step conversion and two kinds of loss: absolute (how many people leave) and relative (what share fails to pass).
- Breakdowns. Compared platforms (iOS vs Android) and acquisition channels (referral vs paid social) at each step to see exactly where the trajectories diverge.
- Difference test. Checked the channel difference with a chi-square test rather than eyeballing averages.
- Visualisation. Built the funnel, a drop-off waterfall and an age heatmap so the bottleneck reads at a glance.
- Data quality. Checked that users move through the steps monotonically (nobody can complete KYC without registering) and that there are no duplicates — or the funnel steps would not reconcile.
- Tooling. Python and pandas/NumPy for the aggregation, SciPy/Statsmodels for the test, Matplotlib/Seaborn for the charts; the generator is seeded, so a run reproduces exactly.
Result
- KYC Complete — largest relative drop-off: 56.6% step conversion (2,781 of 4,917).
- Registration — largest absolute loss: 2,682 users, 73.2% step conv.
- Referral converts +11.7pp better than paid social; iOS beats Android (13.6% vs 11.7%).
- Only 1,269 of 10,000 reach the first transaction (12.7%) — the funnel loses ~87%.
Recommendations
- Rework the KYC UX — progress bar (validated in Project 2), real-time photo hints.
- Simplify registration — A/B test on removing the phone field (largest absolute loss).
- Strengthen referral and reallocate 15% of the paid-social budget.
- Run an Android sprint — QA audit to close the gap with iOS.