Nikita Boyarkin
← Volta Neobank

Extended · project 8 of 23

First-touch, last-touch, linear and Shapley attribution. Shapley (data-driven) reallocates budget and leads with referral; the conclusion is robust to model choice.

Situation
Different attribution models disagree about which channel brings revenue — the comparison runs on one dataset
Task
Own the side-by-side run of competing attribution models to see how much the credited channel depends on model choice
Action
Journey dataset (touch_order × channel × revenue), first-touch, last-touch, linear and Shapley attribution
Result
Referral leads in all four models (€218–264K); Shapley takes referral to €263.7K and drops display to €125.0K
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 — Marketing Attribution

Situation

Different attribution models give different answers about which channel ‘brings’ revenue — we compare them on the same data.

Task

I needed to settle which channel can be credited with revenue, so I owned the side-by-side run of the competing attribution models on one dataset.

Actions

  • Journey dataset (touch_order × channel × revenue).
  • First-touch, last-touch, linear and Shapley attribution.

Result

  • Referral leads in all four models (€218–264K).
  • Shapley reallocates: referral rises to €263.7K, display falls to €125.0K vs ~€180K under the heuristics.
  • Heuristics understate upper funnels and overstate the last click.

Recommendations

  • Reallocate budget per Shapley, not last-touch.
  • Protect referral as the leading channel.
  • Don’t make channel decisions on a single attribution model.

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).

Attribution: four models

Channel revenue under first-touch, last-touch, linear and Shapley attribution (€K). Shapley (data-driven) reallocates budget relative to the heuristics.

0 100 200 300 Referral — First-touch: 228 Paid social — First-touch: 211.4 Organic — First-touch: 203.7 Email — First-touch: 193.7 Display — First-touch: 179.2 Referral — Last-touch: 218.1 Paid social — Last-touch: 220.6 Organic — Last-touch: 200.9 Email — Last-touch: 194.7 Display — Last-touch: 181.8 Referral — Linear: 226 Paid social — Linear: 213.4 Organic — Linear: 203.6 Email — Linear: 194.1 Display — Linear: 178.9 Referral — Shapley: 263.7 Paid social — Shapley: 235.9 Organic — Shapley: 210.3 Email — Shapley: 181.1 Display — Shapley: 125 Referral Paid social Organic Email Display Channel Revenue, €K First-touch Last-touch Linear Shapley
Key takeaways
  • Shapley reallocates budget: referral rises to €263.7K while display falls to €125.0K vs ~€180K under the heuristics.
  • Referral leads in all four models (€218–264K) — the conclusion is robust to model choice.

Volta project map

Volta overview →

Core

Extended

Market & Jobs

RAT v2

Causal