What does MMM estimate, and how?
Media mix modelling estimates how much each channel contributes to a business outcome - revenue, conversions, or leads - from aggregate, time-series data. Rather than following individual users, an MMM regresses periodic outcomes on periodic spend across every channel, alongside non-media drivers such as seasonality, price, promotions, and demand. The fitted coefficients describe how much outcome each channel produces per unit of spend, which is the foundation for everything that follows.
Two transformations make that regression realistic rather than naive. Spend is first passed through an adstock transformation, so each period’s effect carries a decaying contribution from earlier spend instead of treating every week in isolation. It is then shaped by a saturation curve, which bends the response so that early pounds work hard and later pounds buy progressively less. Together they let an MMM separate the true, time-shifted, diminishing-returns contribution of a channel from the raw correlation between its spend and your sales.
Why MMM survives cookie loss
User-level attribution depends on observing individual journeys - pixels, third-party cookies, and mobile identifiers that stitch a click to a conversion. As browsers deprecate cookies, walled gardens close, and privacy regulation tightens, that observability erodes, and attribution increasingly under-reports the channels it can no longer see. MMM does not have this dependency. Because it works entirely from aggregate spend and outcome data, there is no individual to track and nothing to break when identifiers disappear.
That privacy durability is the main reason MMM has returned to favour. It measures every channel on the same aggregate footing - including media that resists tracking, such as connected TV, audio, and offline - and it credits demand-creating channels for impact that lands outside a short attribution window. Where attribution answers “which touchpoint preceded this conversion?”, MMM answers “how much would outcomes change if we moved this budget?” - a question that does not require watching any one person.
A worked paid-media example
Suppose an advertiser runs YouTube alongside paid search and social. Under last-click attribution, YouTube looks weak: viewers rarely click straight through and convert, so the platform’s short window credits search and direct traffic instead. On that basis a planner might cut the channel. An MMM tells a different story. Once adstock carryover is modelled, the model finds that YouTube exposure drives a meaningful share of conversions in the days and weeks after the view - demand that last-click never connected back to it.
With the carryover-adjusted contribution in hand, the picture inverts. YouTube was under-credited, not underperforming, so the sensible move is to increase its budget - say lifting it from £50,000 to £70,000 a quarter - while the model’s saturation curve keeps that increase honest by flagging where additional spend would start to hit diminishing returns. The result is a reallocation grounded in genuine, time-shifted contribution rather than in whichever channel happened to own the final click.
Strengths versus attribution, and how MMM guides budget allocation
MMM and attribution answer different questions and have different blind spots. Attribution is granular and fast but myopic - it only sees what it can track, favours last-click harvesting, and is increasingly degraded by privacy changes. MMM is privacy-durable, covers untrackable media, and captures delayed and diminishing effects, at the cost of slower cadence and a need for sufficient historic variation. The strongest measurement stacks use both: MMM for strategic allocation, attribution for tactical optimisation, and incrementality experiments to validate and calibrate the model’s causal claims.
For media planning, the payoff is forward-looking. Because an MMM fits a response curve per channel, it can estimate the marginal return of the next pound in each one and forecast the outcome of a proposed mix before the spend is committed - true to a forecast-led, marginal approach to planning. That is how MMM becomes a planning tool rather than a backward-looking report. Apply the same logic to your own plan with budget allocation simulator, or see how it shapes a full agency engagement in our guide to paid media forecasting for agencies.
Related terms
- adstock - how advertising impact carries over after exposure, modelled inside the MMM.
- saturation curve - the diminishing-returns shape that caps how much extra spend can buy.
- attribution - the user-level alternative that MMM complements and, under privacy loss, replaces.
- incrementality - the experimental measure used to validate and calibrate an MMM.
Frequently asked questions
What is media mix modelling (MMM)?
Media mix modelling (MMM) is a statistical approach that estimates how much each marketing channel contributes to outcomes such as revenue or conversions. It uses aggregate historical data on spend and results, accounting for adstock carryover and diminishing returns, so it does not depend on tracking individual users.
How is MMM different from attribution?
Attribution follows individual user journeys and assigns credit to the touchpoints it can observe, which makes it vulnerable to cookie loss, walled gardens, and short reporting windows. MMM instead regresses aggregate outcomes on aggregate spend, so it can credit demand-creating channels for delayed impact and measure media that attribution never sees. The two are complementary rather than interchangeable.
Why does MMM survive cookie loss and privacy changes?
MMM never relies on user-level identifiers, pixels, or cross-site tracking. Because it works from aggregate spend and outcome data, the deprecation of third-party cookies, mobile identifiers, and granular click data does not break it. That privacy durability is the main reason MMM has returned to favour for media planning.
What roles do adstock and saturation play in MMM?
Adstock captures how advertising impact carries over after exposure, so a week’s results reflect both current and earlier spend. Saturation curves capture diminishing returns, where each extra pound of spend buys less response as a channel approaches its ceiling. Modelling both is what lets MMM estimate the true, time-shifted contribution of each channel rather than a naive spend-to-sales ratio.
How does MMM inform budget allocation?
Once MMM has fitted a response curve for each channel, it can estimate the marginal return of the next pound in each one. Budget can then be shifted toward channels with the highest incremental return and away from those near saturation, and the expected outcome can be forecast before spend is committed. This makes MMM a planning tool, not just a measurement report.