What is Media Mix Modeling (MMM)?

Media Mix Modeling (MMM) is a statistical modeling technique that marketers use to determine which channels in their marketing mix are driving their sales, acquisitions or other business KPIs. MMM helps brands reallocate budgets to the highest-performing areas of their media mix and cut down on wasted ad spend.

The most important goal of MMM is identifying the incremental return on investment into every media channel in a marketing mix. Incrementality is a measure of what actions would not have happened without a specific intervention.

Consider a brand running ads on Meta. Adding up the users who bought their product after being shown an ad is not the same thing as counting the users who bought their product because of an ad.

Some users who saw the ad would have bought their product anyway, even if the brand had not run their campaign. This is incrementality.

Media Mix Modeling is one of the most popular and practical methods used by brands to estimate the incrementality of each of their marketing channels.

How does Media Mix Modeling work?

Media Mix Modeling uses machine learning and statistical algorithms to find patterns in historical observational data. At a high level, it looks for variation in marketing activity and attempts to line that up with variation in a business KPI.

For example, let’s think about a company with a complex media mix of online and offline channels. On some days, they’ll spend more across their entire media mix than on other days. Sometimes, they’ll spend relatively more on Meta Prospecting ads, and other times they’ll spend relatively more on TikTok Retargeting ads.

Media Mix Modeling identifies the patterns in these variations: on days when this brand spent relatively more on Meta Prospecting, how many additional new customers did they generate? On weeks when they scaled investment across their media mix, did they see an increase in revenue?

If there is no observed increase in the brand’s core KPI during periods of increased marketing activity (after accounting for adstock and other factors), we’ll know that spend wasn’t very incremental.

MMM does this type of analysis over and over again for all marketing channels across all of the days they were active, until it can provide estimates of incrementality across the entire media mix.

Media Mix Modeling was first developed in a context where tracking was near impossible. The idea was to use aggregate, not individual-level, data to find the statistical relationships between variation in marketing activity and variation in business KPIs.

Modern MMMs still work with aggregated data and are not reliant on tracking pixels or cookies. This makes them an attractive alternative to tracking methods that have been impacted by GDPR, iOS14, and increased privacy regulations.

What types of problems does MMM solve?

Every attribution method has unique capabilities and limitations, and Media Mix Modeling is no different. MMM is typically not suited for daily optimizations at the creative or audience level due to the signal it requires. Where MMMs thrive is in powering decisions about overall marketing strategy and budget allocation.

What categories of strategic-level questions can MMM help solve? Here are a few examples:

• Forecasting and Planning: How many new customers will our current media mix drive over the next 12 months if we hold budgets steady?
• Budget Optimization: If my quarterly revenue goal is $7 million, what is the ideal spend across every channel in my media mix to achieve this?
• Scenario Analysis: My business needs to operate at a ROAS greater than 3.0. What is the most we can spend while staying above this level and which channels should we allocate this spend to?

MMMs can answer tons of important questions for a business and make meaningful impacts on performance. In terms of output, these are a few categories you should expect from a good model:

• ROI, marginal ROI and CPA estimates for every media channel: These allow marketers to make budget decisions by telling them which channel(s) they can invest their “next dollar” into most effectively.
• Communication of uncertainty: For every estimate that a model produces, the point estimate should be accompanied by an uncertainty interval expressing the amount of uncertainty in the estimate. This helps marketers make effective decisions about where to invest their dollars and makes them aware of the magnitude of risk they might be taking with these decisions.
• Forecasts of future sales based on budget inputs: A good MMM should be able to predict how much revenue or customer acquisitions will be produced by a given marketing budget into the future, and provide uncertainty intervals expressing the uncertainty in the forecast based on these inputs.
• Optimized media mix based on current performance: A good MMM should also be able to produce realistic and optimal budgets for every channel in a media mix based on constraints provided by the marketing team.

These are powerful outputs, but for many marketing teams it’s not enough to rely on a single tool. For this reason, we recommend combining MMM with other attribution methods in a process called “triangulation”.

This modern way of thinking about measurement unites digital tracking (also known as multi touch attribution or MTA), Media Mix Modeling (MMM), and testing/conversion lift studies (CLS).

How does this work in practice? You might use MTA for directional data and daily platform optimizations, MMM for forecasting and budget allocation decisions and conversion lift studies to calibrate the accuracy of your MMM. Combining these methods together will get you closer to finding true incrementality in your media mix.

The History of Media Mix Modeling

MMM has historical roots that trace back to the 1960’s when large CPG companies pioneered the use of models to evaluate the impact of their ad campaigns. This methodology provided empirical evidence and insights into their most profitable channels.

During this time, the vast majority of marketing spend was allocated towards broadcast channels like linear TV and radio. Advertising spots were bought via “upfronts”, where brands would buy all of their media for the next 6-12 months.

Legacy MMM vendors (and a few to this day) tailored their methodology around these upfront media buys. They ran their model a few times a year, assessed the relative efficacy of different marketing channels, and then made a recommendation about how much a brand should buy in upfronts.

Not surprisingly, the old way of doing MMM developed 50+ years ago doesn’t match the pace of modern marketing activity. Fortunately, modern marketing mix modeling methods address many of the limitations of these legacy approaches.

Modern MMM platforms are designed for in-flight media optimization. They’re fast - producing actionable insights in near real-time - and dynamic to account for day-to-day changes in channel performance. Modern MMMs are also verifiable: their models are not black boxes, but rather, empower marketers to verify their accuracy.

In short, they keep pace with today’s marketers and provide tools to validate all of their output.

Is MMM Right For Your Brand?

Media Mix Modeling requires a significant investment of resources, and for that reason, is not a perfect fit for every brand. If you’re thinking of undertaking a project, there are a few things you should consider.

The first is the number of marketing channels you operate in. If your company’s marketing strategy is concentrated in just one or two channels, like Meta and Google SEM, in-platform reporting and lift tests might be enough and you don’t need to overcomplicate your measurement strategy. MMM’s strength comes from analyzing and interpreting interactions across a broad spectrum of channels. With a limited channel mix, the potential insights that MMM can yield are constrained.

The scale of your marketing budget is another critical factor. While MMM can be applied to any size of marketing budget, its true value is shown with larger budgets. With less than $5 million per year in media spend, it’s often hard to make a project successful given how complicated MMM is and how easy it is to get wrong.

A third, and often overlooked, consideration is the willingness of your organization to act on the insights provided by MMM. It’s not uncommon for companies to do MMM, get some valuable insights, and be unable or unwilling to implement the recommended changes.

If your marketing team is not prepared to pivot strategies, reallocate budgets, or experiment with new channels based on the insights from MMM, then the exercise becomes an academic one rather than a transformative business initiative.

MMM requires a lot of assumptions and deep knowledge of sophisticated modeling techniques to get right. It’s much harder to build an accurate MMM than it might seem at first glance. Because of this, your team should be ready to answer a few critical questions before starting a project:

• How will we know if the insights from the MMM are accurate or not?
• How will we use probabilistic results that may change as we get new data?
• What happens if the results from the MMM conflict with other data sources?

So who is MMM right for? If your brand is making a meaningful investment into a diverse media mix that includes hard-to-measure channels, it might be time to consider a project.

Final thoughts

Media Mix Modeling is a powerful statistical modeling technique that looks for variation in marketing activity and attempts to line that up with variation in a business KPI. It’s not reliant on digital tracking methods, but rather, uses aggregated data to glean insights into marketing performance.

MMM has come a long way since the legacy methods developed 50+ years ago. Modern solutions are helping marketers find true incrementality in their media mix while encouraging them to verify the accuracy of their models.

Is MMM right for your brand? If you’re making a significant investment into a media mix that includes hard-to-measure channels and have a marketing team that’s ready to take action on the insights from MMM, it might be time to think about a project.

Just remember: MMM is a single (but important) layer on an attribution stack. It should be used in conjunction with other attribution methods to find true incrementality in your marketing activities.

Want to keep reading? Follow Michael Kaminsky on LinkedIn for essays on marketing effectiveness and check out Recast to see what a modern MMM platform looks like.