Mix Modelling Is No Longer Only for Giants

Marketing mix modelling uses aggregate time-series data — spend, sales, price, seasonality — to estimate how much each input contributes to the outcome. Because it never touches individual-level data, it is untouched by cookie deprecation, consent rates or app tracking restrictions. That is why it has returned to fashion.

What you need before starting

The two concepts that do the work

Adstock captures the fact that advertising keeps working after the week it ran. Saturation captures diminishing returns: the tenth thousand pounds in a channel does less than the first. Together they turn a naive regression into something that can answer the only question that matters, which is where the next unit of budget should go.

A model without saturation curves will always recommend spending more on whatever performed best last quarter.

Where it goes wrong

The classic failure is correlated spend: if you always raise every channel before Christmas, the model cannot separate them. Deliberately varying budgets — staggering increases, occasionally pausing a channel — creates the variation the model needs. Spend patterns that never change produce confident nonsense.

Use it for the right decisions

MMM operates at weekly aggregate resolution. It can tell you that one channel is saturated and another is under-funded. It cannot tell you which creative to pause on Tuesday. Pair it with platform data for daily work and experiments for causal checks, and treat the three as complements rather than competitors.

Refresh cadence

Quarterly refreshes are enough for most businesses. Monthly re-fits tend to produce noise that gets over-interpreted, and the budget decisions the model informs are not made weekly anyway.