What is marketing mix modeling?

Marketing mix modeling, usually shortened to MMM, is a statistical approach that measures the effect of marketing on business outcomes from aggregated data: weekly or daily spend per channel on one side, sales, orders or applications on the other. It needs no cookies, no pixels and no user-level tracking, which is why it works for TV, radio and out-of-home as well as for search and social.

The model accounts for the things that move demand regardless of marketing: seasonality, price and promotions, and large external events. What remains after those are explained is the part each marketing activity caused, with diminishing returns as spend grows and effects that can land weeks after the money was spent.

The model starts from what the company has done historically (two years of history is the comfortable floor, six months the minimum) and estimates a response curve for every marketing activity: how much extra sales each level of spend produced, and where the returns flatten out. Carryover, called adstock, captures effects that arrive later. Control variables such as seasonality and price explain the demand that would have happened anyway, the baseline.

Modern MMM is Bayesian. Instead of starting from a blank slate, the model starts from what is already known about each channel, encoded as priors, and lets the data update that starting point. The result is a posterior for every parameter: a most likely value and a range, never a single number pretending to be certain.

The outputs are the ones budget decisions need. Channel contribution says what each activity added beyond the baseline. The saturation curve says where the next unit of spend in a channel still pays and where it has flattened. Marginal ROAS puts a number on that next unit. From those, scenario planning answers the two questions every budget cycle comes down to: are we spending the right amount in total, and is it going to the right places?

At Odins the model is built and retrained by our team. Data flows in daily, the model is retrained monthly, and the platform is open any time for reporting, scenarios and forecasts. The monthly retrain is as often as a budget decision changes; retraining more often on the same weekly data would not add anything.

MMM is not attribution. Attribution credits the touchpoint before a conversion; MMM measures what spend caused in aggregate, including the sales a channel never gets a click for. The two answer different questions, and the platforms' own tracking remains the best tool for optimizing inside each platform. It is just not a currency that works across platforms.

MMM also works at the level the data supports. One national weekly series per market can carry roughly ten activity types with good priors; it cannot reliably say anything about individual ads or ad groups. A vendor promising ad-level recommendations from an MMM is promising more than the data can hold.

• Bayesian marketing mix modeling: A model that starts from what you already know and lets the data correct it, with a range on every answer.
• Saturation curve: The response curve per channel: how sales respond to spend, and where the next unit stops paying.
• Adstock: The carryover effect: the part of this week's advertising that lands in the weeks that follow.
• Baseline (base sales): The sales that would have happened without paid marketing. Everything the model credits to a channel is measured above it.
• Channel contribution: What each channel added to sales above the baseline over a period. The decomposition behind every ROAS figure.
• Marginal ROAS: What the next unit of spend in a channel earns at today's level. The number a budget change actually asks for.
• Attribution: Crediting conversions to the touchpoints that preceded them. Great inside a platform, not a currency across them.

Go one level deeper: The complete guide to marketing mix modeling. How the Odins model works. Or go back to all terms in the glossary.