How much data do you need for marketing mix modelling?

At minimum, two years of weekly data is recommended to capture seasonal patterns and have enough variation for reliable regression results. Three or more years is better. Fewer than 100 weekly observations typically produces models with wide confidence intervals that are hard to trust for budget decisions.

Can small businesses use marketing mix modelling?

Traditional MMM requires significant data and statistical expertise, making it more suited to mid-size and large companies. However, simplified versions using tools like Google's open-source Meridian or Meta's Robyn can work for smaller teams with at least two years of spend data and a willingness to learn the fundamentals.

How often should you refresh an MMM?

Quarterly updates are standard practice. This cadence captures recent shifts in channel performance and market conditions while maintaining enough new data for meaningful model updates. Major business changes like entering a new market or launching a new product category may warrant an immediate model refresh.

Marketing IntelligenceAdvanced5 min read

What Is Marketing Mix Modelling?

Marketing mix modelling uses statistical analysis to measure how each marketing channel contributes to business outcomes. Learn how MMM works.

Key Takeaways

Marketing mix modelling uses regression analysis on aggregate data to quantify each channel's impact on revenue.
MMM does not rely on user-level tracking, making it privacy-resilient.
It requires significant historical data and expertise to build reliable models.

What marketing mix modelling is

Marketing mix modelling is a statistical technique that analyses historical data to determine how different marketing inputs, such as TV spend, digital advertising, promotions, and pricing, contribute to business outcomes like revenue or unit sales. Unlike attribution modelling, which tracks individual user journeys, MMM works with aggregate data. It uses regression analysis to isolate the effect of each variable while controlling for external factors like seasonality and economic conditions.

Why MMM is gaining renewed attention

As privacy regulations and cookie restrictions make user-level tracking increasingly difficult, MMM offers an alternative that does not depend on tracking individual consumers. Tech companies including Google and Meta have released open-source MMM tools, making the technique more accessible. For marketers who spend across both online and offline channels, MMM provides a unified view that digital attribution models cannot, measuring the impact of billboards and radio alongside search ads.

How MMM is built

Building an MMM requires two or more years of weekly data covering marketing spend by channel, sales or revenue figures, and external variables like holidays, competitor activity, and economic indicators. A data scientist fits a regression model that explains sales variation based on these inputs. The model outputs channel-level contribution estimates and diminishing return curves that show the optimal spend level for each channel.

Limitations and best practices

MMM is slow to update because it relies on historical trends, making it poorly suited for real-time optimisation. It also requires substantial data volumes, which can be challenging for smaller companies or those in newer markets across Africa with limited historical baselines. Best practice is to use MMM for strategic budget allocation decisions and complement it with attribution modelling and incrementality testing for tactical channel-level optimisation.

What Is Attribution Modelling?4 min · Intermediate What Is Multi-Touch Attribution?5 min · Advanced What Is Incrementality Testing?5 min · Advanced