← Volta Neobank

Volta — A/B Testing (KYC progress bar)

Stack
Pythonpandas / NumPySciPy / StatsmodelsMatplotlib / Seaborn
Impact
  • +6.24pp KYC (Z=6.35, p<0.0001), 95% CI [+4.26%, +8.16%], exceeds +5pp MDE
  • CUPED + AA-test (type-I=0.050) + Bonferroni (4/11 survive)
  • €716K/yr business impact (48× ROI)

Volta — A/B Testing (KYC progress bar)

Business Context

The second project in the validate loop: does a progress bar in KYC fix the problem found in the funnel? A ship-gated methodology protects against rolling out statistically-significant but business-insignificant changes.

Hypothesis

A KYC progress bar raises KYC completion conversion by at least +5pp (MDE) with no SRM.

Data & Method

  • Sample size calculation for MDE = +5pp.
  • SRM check (sample ratio mismatch): p = 1.00 → balanced samples.
  • Bootstrap CI + multiple-comparison correction (Bonferroni / Holm / BH).
  • AA-test under H₀: type-I error = 0.050 (methodology calibrated).
  • CUPED (control-only θ): variance reduction, sensitivity at MDE.
  • Ship-gate: p<0.05 ∧ lift ≥ MDE ∧ no SRM → ship.

Insight

  • Control 55.8% → treatment 62.1%, +6.24pp, 95% CI [+4.26%, +8.16%].
  • Exceeds MDE (+5pp), p<0.0001 (Z = 6.35).
  • 9/11 naively-significant segments → 4/11 after Bonferroni (correction removes false positives).

Impact

  • KYC conversion +6.24pp → business impact €716K/yr (48× ROI on €15K dev cost).
  • The methodology (CUPED + AA-test + Bonferroni + ship-gate) is reusable for future experiments.

Documentation

Other Volta subprojects