attribution-model-analysislisted

# Attribution Model Analysis This skill turns a raw touchpoint log into a defensible read of which channels actually contribute to conversions, by computing several attribution models side by side and interpreting the differences between them. ## Outcome contract - **Outcome**: a channel by model comparison table plus a reasoned recommendation of one reporting model, always bracketed by first-touch and last-touch as the bounds of the honest uncertainty. - **Done when**: every model column reconciles to the same total conversion value, the recommendation cites observable properties of the dataset, and the limits of attribution (no incrementality claim) are stated in the output. - **Evidence**: the user's touchpoint log, the reconstructed per-conversion paths, and per-path share sums verified to equal 1. ## When to use this skill Reach for it in these situations: - **Channel credit disputes.** Paid social claims the conversions, paid search claims the same conversions, and the sum of platform-reported conversions is larger than the number of orders or deals that actually closed. - **Last-click looks suspicious.** Branded search, direct, and email absorb almost all credit, while the channels that generate demand show a poor return. That is the classic signature of a single-touch model, not a real result. - **Choosing a reporting model.** The team needs one model for the monthly dashboard and has to justify the choice to finance or to a client. - **Sanit