Growth Engineering Playbook · Wave 5
How much of your sales did marketing actually cause?
Attribution splits credit among the touches it can see; a marketing mix model answers the bigger question — of all the sales that happened, how many did advertising drive, and how many were the base that was coming anyway. It regresses weekly sales on spend, but with the two transforms that make ad spend behave like ad spend: adstock (carryover that decays over weeks) and saturation (diminishing returns). This fits both from two years of Northstar Outfitters data, decomposes sales into base + channels, and derives each channel's ROI. It's correlational, not causal — so it says, loudly, to validate against holdouts. Deterministic, client-side.
Fitting model…
Sales decomposition
Two years of sales split into the organic base and each channel's modelled contribution.
Channel response
One channel's saturation curve + carryover.
Model fit — actual vs predicted weekly sales
The model reconstructs history. A high R² means the decomposition is worth reading; a low one means don't.
Channel breakdown
ROI is modelled incremental revenue ÷ spend. Carryover is the extra effect from adstock — how much of a channel's punch lands in later weeks.