Marketing mix modelling (MMM) uses historical data and statistics to estimate how much each marketing channel contributed to your sales. It gives you a single model that covers all channels at once, without relying on cookies or pixel tracking.
Every marketing director faces the same problem. You spend money across TV, paid search, social media and out-of-home. Sales go up. Sales go down. But which channel actually caused the movement? MMM was built to answer that question.
The method has been used by large consumer goods companies since the 1980s. What has changed recently is that better computing power and open-source tools have made it accessible to mid-sized brands too.
The basic idea: separating signal from noise
A marketing mix model is a regression equation. Regression is a statistical technique that finds the relationship between one thing you want to explain (your sales) and a set of things that might explain it (your media spend, price, promotions, seasonality and so on).
The model looks at weeks or months of data and asks: when spend on a particular channel went up, did sales tend to go up too? It controls for other factors, such as a competitor launching a promotion or a bank holiday boosting footfall, so you get a cleaner read on each channel.
What goes into the model
The inputs are usually weekly or monthly figures. A typical dataset includes media spend by channel, revenue or sales volume, price and any promotional discounts, seasonality indicators, and macroeconomic factors such as consumer confidence.
- Media spend by channel (TV, paid search, paid social, display, radio, OOH)
- Revenue or sales units, split by product or region if needed
- Price and promotional discount data
- Seasonality variables such as month of year or school holidays
- External factors such as weather, competitor activity, or economic indicators
The more complete this data is, and the longer the time period it covers, the more reliable the model becomes. Most projects use at least two years of weekly data.
How the model separates each channel
The model assigns a coefficient to each channel. A coefficient is a number that shows how much your sales change, on average, for every additional pound spent on that channel, while everything else stays the same. A higher coefficient means the channel is more efficient.
These coefficients let you calculate the return on investment (ROI) for each channel. If TV has a coefficient of 2.5 and paid search has a coefficient of 4.1, it tells you that paid search delivered more sales per pound spent over the period you modelled.
The model does not track individuals. It works at an aggregate level, looking at total spend and total sales across a market. This makes MMM privacy-safe and unaffected by cookie deprecation.
Adstock and saturation: making the model realistic
Two adjustments make the model more accurate. The first is adstock, which captures the fact that advertising does not stop working the moment you switch it off. A TV ad seen on Monday still influences a purchase made on Friday. The model applies a decay rate to account for this carry-over effect.
The second adjustment is saturation. Doubling your spend rarely doubles your sales. The model applies a curve that flattens out at higher spend levels, reflecting the diminishing returns that come from overexposure.
What you do with the results
Once the model is built, you can run budget optimisation scenarios. If your total budget is fixed, the optimiser finds the allocation across channels that is expected to produce the highest sales. You can also model what happens if your budget increases or decreases by 20 percent.
Most teams run a new model each quarter or each year, updating it with fresh data. This keeps the coefficients current and lets you track whether channel efficiency is improving or declining over time.
How is MMM different from last-click attribution?
Last-click attribution gives all the credit to the final touchpoint before a conversion, usually a search ad. MMM looks at the whole picture across all channels, including offline ones, and measures contribution rather than credit.
Does MMM work for companies that sell offline as well as online?
Yes. MMM was originally designed for offline sales and works just as well for businesses with physical stores, phone sales, or mixed channels. You simply use total revenue as the outcome variable.
How many channels can a marketing mix model cover at once?
There is no fixed limit, but most models cover between 5 and 15 channels. Adding more channels requires more data to keep the estimates reliable. Your modeller will advise on the right level of detail for your dataset.