Nikita Boyarkin
← Volta Neobank

RAT v2 · project 22 of 23

Three-arm win-back: human +5.34pp and light-touch +2.77pp vs control. Light-touch pays off at 30–90d (ROI 2.54 / 1.26), human ROI 0.40 — kill as a mass channel.

Situation
Closing project of the validation round: an earlier part tied senior dormancy to a UX barrier but never tested reactivation
Task
Own the check of whether assisted reactivation actually brings dormant senior users back and whether it pays for itself
Action
Three-arm win-back, four dormancy buckets, z-tests by arm and bucket with Holm correction, ROI and targeting boundary
Result
human +5.34pp, light-touch +2.77pp; light-touch pays off at 30–60d (ROI 2.54) and 60–90d (1.26), human overall 0.40
Stack
Pythonpandas / NumPySciPy / Statsmodelsscikit-learnMatplotlib / Seabornuv + ruff
Source View on GitHub →
On this page
  1. Situation
  2. Task
  3. Actions
  4. Result
  5. Recommendations
  6. Documentation

Volta — Dormant 45+ Win-back

Situation

The last RAT v2 project: Project 13 argued that 45+ dormancy is a UX barrier. We test whether assisted reactivation recovers these dormant users and whether it pays off.

Task

I owned the closing check of the validation round: whether assisted reactivation actually brings dormant senior users back, and whether that effort pays for itself.

Actions

  • Three-arm randomized win-back: automated / light-touch / human.
  • Four dormancy buckets; z-tests by arm and bucket with Holm correction.
  • ROI per 10K treated, targeting boundary, break-even LTV.

Result

  • Mechanism confirmed: human +5.34pp and light-touch +2.77pp vs control (both significant); strongest at 30–60d (human +11.37pp).
  • Economics refine it: light-touch pays off at 30–60d (ROI 2.54) and 60–90d (1.26); human loses money everywhere (overall 0.40).
  • At 180d+ the effect nearly vanishes.

Recommendations

  • Launch light-touch win-back targeting 30–90 day dormant users (ship narrowly).
  • Human calls are an escalation for high-value/high-balance, not a mass channel (kill).
  • Don’t touch 180d+ — the economics are negative.

Documentation

Charts

Source: github.com/NikitaBoyarkin/volta-banking — 22 projects; figures recomputed from the repo's own datasets (data/*.csv) via its analysis scripts. Funnel counts from data/volta_funnel_data.csv (10,000 users); A/B, retention, segmentation, churn, RFM, CLV, attribution, anomalies, spend, support, NPS, JTBD, unit economics, premium, KYC deep-dive, referral, assisted CAC, FX sourcing, premium offers, anchor CAC and dormant win-back follow the published project narrative (README + part pages).

Dormant 45+ win-back: reactivation by arm × depth

60-day reactivation (%) for three arms (control / light-touch / human) across four dormancy buckets. Assisted reactivation works, strongest for the recently dormant.

0 5 10 15 20 30-60d — Control (automated): 8.47 8.47 60-90d — Control (automated): 5.7 5.7 90-180d — Control (automated): 3 3 180d+ — Control (automated): 1.57 1.57 30-60d — Light-touch: 14.43 14.43 60-90d — Light-touch: 9.07 9.07 90-180d — Light-touch: 4.8 4.8 180d+ — Light-touch: 1.5 1.5 30-60d — Human-assisted: 19.83 19.83 60-90d — Human-assisted: 12.07 12.07 90-180d — Human-assisted: 5.93 5.93 180d+ — Human-assisted: 2.27 2.27 30-60d 60-90d 90-180d 180d+ Dormancy depth Reactivation, % Control (automated) Light-touch Human-assisted
Key takeaways
  • Assisted reactivation works: human +5.34 pp and light-touch +2.77 pp vs control overall (both significant).
  • Strongest at 30–60d (human +11.37 pp); at 180d+ the effect nearly vanishes.

Dormant 45+ win-back: ROI by arm × depth

ROI (recovered LTV / incremental cost) per 10,000 treated by arm and dormancy depth. ROI=1 is break-even.

-1 0 1 2 3 30-60d — Light-touch: 2.54 2.54 60-90d — Light-touch: 1.26 1.26 90-180d — Light-touch: 0.56 0.56 180d+ — Light-touch: -0.02 -0.02 30-60d — Human-assisted: 0.92 0.92 60-90d — Human-assisted: 0.45 0.45 90-180d — Human-assisted: 0.17 0.17 180d+ — Human-assisted: 0.03 0.03 30-60d 60-90d 90-180d 180d+ Dormancy depth ROI Light-touch Human-assisted
Key takeaways
  • Light-touch pays off at 30–60d (ROI 2.54) and 60–90d (ROI 1.26), beyond that it does not; human never pays off (overall 0.40).
  • The target track is light-touch only at 30–90 days; human is an escalation for high-value/high-balance.

Volta project map

Volta overview →

Core

Extended

Market & Jobs

RAT v2

Causal