SAMPLE REPORT
A weekly decision sheet, not a dashboard export.
Built from Google's public GA4 ecommerce sample. Figures are real model outputs on that dataset, labeled as such. Your PDF uses your conversion definition.
PULSAR ANALYTICS
Weekly attribution · sample property
Public dataset · 90-day lookback · key event: purchase
RUN 00-PUBLIC
01 · Decision sheet
Planning model this week: Markov removal. Last-click over-credits Paid Search by 8 pp and Direct by 10 pp. Email and Referral are the channels last-click is stealing from.
Move
Do not cut Email on last-click ROAS. It is +6 to +8 pp under path models.
Hold
Paid Search is still the largest path contributor — just not 34%.
Human
Direct share collapsed under removal. Confirm brand-query capture before celebrating.
02 · Comparison
| Channel | Last-click | Markov | Shapley | Δ vs LC |
|---|---|---|---|---|
| Paid Search | 34% | 26% | 24% | -8 pp |
| Direct | 22% | 12% | 11% | -10 pp |
| Organic Search | 14% | 18% | 19% | +4 pp |
| Paid Social | 12% | 11% | 10% | -1 pp |
| 8% | 14% | 16% | +6 pp | |
| Referral | 5% | 10% | 11% | +5 pp |
| Organic Social | 3% | 5% | 5% | +2 pp |
| Affiliates | 2% | 4% | 4% | +2 pp |
03 · Markov removal
Removing Email from converting paths drops reconstructed conversions 14%. Removing Paid Search drops 26% — material, but 8 pp below last-click. Direct removal barely moves the count: it is a closer, not a cause.
04 · Shapley interval
Monte Carlo, 5,000 permutations, 90% interval. Email phi 16% [13, 19]. Paid Search 24% [21, 27]. Intervals do not overlap last-click for Email or Direct.
06 · Recommended test
If Paid Search vs Email disagreement stays above 6 pp for two more weeks, pause a non-brand Search cluster in one geo for 14 days. Do not read that as MMM. It is a cheap incrementality check on an expensive argument.
Appendix omitted in this sample: conversion definition, channel map, data-quality score, BQ bytes billed. LLM commentary is verifier-checked against this table only.
INSIDE EVERY PDF
01
Decision sheet
Recommended planning model this week, top 3 budget implications, anomalies that need a human.
02
Comparison heatmap
Channel × six models, delta versus last-click. Shares reconcile to conversion count.
03
Markov removal
Transition highlights and what conversions disappear if a channel is removed.
04
Shapley with uncertainty
Channel phi and a 90% interval. Exact if the coalition is small; Monte Carlo otherwise.
05
Stability
Week-over-week rank correlation versus the last eight weeks. Share swings flagged.
06
Recommended test
A geo or pause experiment if disagreement is expensive enough to be worth it.
PUBLIC GA4 SAMPLE
Last-click vs path models
- Last-click
- Markov
- Shapley
Paid Search
-8 pp
Direct
-10 pp
Organic Search
+4 pp
Paid Social
-1 pp
+6 pp
Referral
+5 pp
Organic Social
+2 pp
Affiliates
+2 pp
Share of credited conversions. Negative pp means last-click is over-crediting that channel versus Markov removal on Google's public ecommerce sample.