Reference library · Media planning
Marketing Mix Modeling (MMM)
Marketing Mix Modeling (MMM) is a statistical technique that regresses a business outcome, usually sales or revenue, against historical marketing spend by channel and control variables like price, distribution and seasonality, to estimate each channel's incremental contribution and its response curve. The output shows how much of past sales each channel actually drove, and where each channel sits on its curve of diminishing returns. Planners use it to reallocate budget away from channels showing saturation and toward channels with remaining headroom.
Worked example
A brewery in Poland runs an MMM on 24 months of weekly data covering a €2.1M quarterly budget split across TV, digital and promotions. The model decomposes sales into: baseline (distribution, price, seasonality) 62%, TV 18%, digital 12%, promotions 8% (62+18+12+8=100%). The TV response curve flattens above €150k/week while digital is still near-linear up to €80k/week, so the planner shifts €400k from TV to digital for the next quarter at the same total budget, and the model forecasts +3.2% incremental sales from that reallocation.
How it is used
MMM runs on aggregated time-series data (weekly or monthly spend and sales, typically 2-3 years of history) rather than individual exposure like a TRP or reach model; it separates baseline sales from factors outside marketing's control from incremental sales attributable to each channel, then fits a response curve per channel to capture diminishing returns. Planners rerun the model quarterly or after a major spend shift and use the resulting contribution and elasticity estimates to move budget from saturated channels toward channels still on the linear part of their curve. The common failure is treating a single model run as fixed truth and reallocating on it without checking collinearity between channels that ran simultaneously, such as TV and digital flighted together, which inflates or steals credit between them.
The common mistake
Treating one model run's channel splits as permanent instead of re-estimating each cycle, since contributions shift with seasonality, competitive activity and each channel's own saturation point.