Media mix modeling (MMM)

Attribution follows individual journeys. Media mix modeling ignores individuals and looks at aggregates: weekly sales against weekly spend by channel, with seasonality, promotions, pricing, and trend accounted for. The model estimates each channel contribution and how quickly returns diminish as spend rises.

Because it works on aggregates, MMM does not care about cookies, consent rates, or cross-device journeys. That is its strength. Its weakness is data appetite. A model needs enough history and enough variation in spend to separate one channel from another, and channels that always move together cannot be told apart by any regression.

Open-source tools such as Google Meridian and Meta Robyn have lowered the cost of building a model, but not the cost of feeding one. Two years of clean weekly data, deliberate spend variation, and at least one experiment to calibrate against are what separate a useful model from a confident-looking chart.

For most brands spending under a few hundred thousand dollars a month, disciplined incrementality tests on the biggest channels answer the budget question sooner and more cheaply. MMM earns its place when the channel mix is wide enough that testing each one in turn would take years.