Marketing mix modeling (MMM)

MMM looks at the big picture. Instead of tracking individuals, it takes weekly or daily totals, such as spend by channel, sales, prices, promotions, weather, holidays and competitor activity, and uses regression-style models to estimate how much each factor drove results.

Because it needs no user-level data, MMM is privacy-friendly and works across offline and online channels, including TV, radio, print and retail media. It captures effects that clicks miss, like the long-term brand building of TV. Its limitations are that it needs a lot of history, reacts slowly, struggles to separate channels that always move together and can be sensitive to modelling choices.

MMM has had a revival as signal loss weakens multi-touch attribution. Open-source tools have made it more accessible, including Meta's Robyn and Google's Meridian. Best practice is to calibrate MMM with incrementality experiments such as lift studies and geo tests, so the model's estimates are anchored to causal evidence.

Think of it like this

MMM is like working out which ingredients make a cake taste good by comparing hundreds of past bakes, rather than watching each bite being eaten.

An example

A packaged-food company in India runs MMM on three years of weekly data and finds TV drives 30% of incremental sales, digital video 15%, and trade promotions 25%, then reallocates its budget for the next year.