What is marketing mix modeling?

The wider frame

Marketing mix modeling is the broader parent of media mix modeling. Media mix models estimate the contribution of media channels; marketing mix models include the levers that are not media at all — price, promotion, distribution, product changes, and the competitive and macroeconomic environment.

The distinction matters because those non-media levers are usually larger than the media ones. A model that omits a price cut and a distribution gain will attribute their effect to whatever advertising happened to run in the same weeks, and produce a very flattering and completely wrong read on media.

What it typically includes

• Base and incremental. The volume the business would have done anyway, separated from what marketing added.
• Media contribution by channel, with diminishing returns curves.
• Price and promotion elasticity — usually the strongest short-term effects in the model.
• Distribution and availability, which dominate in physical retail.
• Seasonality and external factors, so weather, holidays and category shifts are not credited to a campaign.

How to read the output honestly

Two disciplines separate useful models from decorative ones.

Report the uncertainty. A contribution estimate is a range. Presenting the midpoint alone converts a statistical result into a false certainty, and decisions get made on the third decimal place of a number with a 40% confidence interval.

Validate out of sample. A model that fits history perfectly and has never predicted a period it did not see is not evidence of anything. Holding back recent periods and testing the prediction is the minimum standard, and it is skipped more often than not.