The Beginner’s Guide To Marketing Mix Modeling (MMM)
Marketing Mix Modeling is a black box to a lot of people, but just like incrementality testing, it’s increasingly important in a world where you can’t always track your users. Mike Taylor has deep experience in MMM and laid it all out for us in this beginner’s guide.
You're a business owner who has been pouring money into various advertising channels - Google ads, social media promotions on platforms like Facebook and TikTok, and so on. Yet, you're left scratching your head, trying to figure out which of these platforms is giving you the best return on your investment. Here is where Marketing Mix Modeling (MMM) comes to your rescue.
Simply put, MMM is like a marketing detective that sifts through piles of historical sales and marketing data and uses statistical analysis to discover which elements of your marketing mix - be it advertising, price changes, or promotions - are truly driving your sales. It uses statistical analysis to match spikes and dips in sales to events and actions in marketing without having to track individual user-level data.
In this article, we’ll explain the ins and outs of MMM so you know when's the right time for your business to invest in one and what to watch out for. This is a beginner’s guide, so we’re explaining it from a business perspective, with no jargon or mathematical formulas.
The Basics of MMM
To grasp how MMM works, imagine you run an ecommerce shoe brand. You spend on Facebook, Instagram, and Google ads but aren’t sure which drives the most sales. Google claims a lot of sales, but you see organic sales go up when you increase spend on Facebook. You also see more searches for your brand, which then Google claims credit for.
Building an MMM from historical data on your marketing spend and sales, you get a statistical model showing how much each channel really contributes, independent of what they report. When you compare your marketing attribution results to your MMM, you might find that Facebook and Instagram ads really are underreporting how many sales they’ve really driven! Without that knowledge, you might have incorrectly concluded that they weren’t performing and put too much money into Google, which may be less incremental.
MMM also reveals when spending more won’t boost sales—the dreaded point of diminishing returns. So when you double your budget, you know how many sales to expect from that increase in spend. You’ll find that some channels have a lower ceiling before they get inefficient. Armed with this knowledge, you can make data-backed decisions to optimize your marketing. What’s more, is that you get all this insight without invading user privacy: it uses aggregated data rather than pixels or cookie-based tracking, which means it also works for offline channels.
MMM in Action: A Case Study
One of the most open and shut cases of MMM I worked on was for an ecommerce client in a similar position. They ran social ads, PPC, and email campaigns to promote their online store but had questions about whether Google was really incremental, and suspected their TikTok ads were driving more value than was being measured. After building an MMM, we uncovered the following:
- Google ads drove only 15% of sales—not the 40% reported. We reduced spend here and took it in-house, saving money on an external agency.
- Facebook was their powerhouse, driving 60% of sales. Within Facebook, their retargeting was only 20% incremental, so we reallocated the budget into prospecting.
- Email newsletters were stealing too much credit, claiming 25% of sales but really contributing more like 5%. This is typical for channels that target people who are already very likely to buy.
- TikTok ads, as the client suspected, were a secret weapon. Much of the value claimed by Google was really branded searches driven from TikTok videos. We doubled this budget.
Within months of optimizing based on MMM insights, marketing ROI increased over 35%. MMM allowed us to redistribute budget to the most effective channels, tightening targeting and boosting success.
MMM and Privacy
In today’s data privacy-focused world, MMM - just like incrementality testing - offers marketers an analytical solution that doesn’t compromise user privacy. Unlike tracking methods that follow individuals across websites, MMM uses statistics on aggregate data to uncover what’s driving sales.
As privacy regulations like GDPR and iOS14 restrict tracking, MMM has become an essential piece of the puzzle for marketers to understand their impact without infringing on privacy. For instance, if iOS14 impacts your Facebook ad data, MMM can still show Facebook’s overall influence on revenue based on your total Facebook spend and sales. You can then adjust budgets accordingly.
This great strength is also a weakness: because it relies on aggregate data, it can’t provide as much detail at the campaign or creative level as you get from platform reports. Rather, MMM helps set your marketing direction by revealing how each channel impacts the other and your bottom line. For example, MMM may show that for each dollar spent on search ads, you can reduce display ad spend by 50 cents while maintaining revenue. You can strategically allocate your budget for the best total return. This is why strategic brands often combine MMM with an incrementality modeling solution - to get both aggregate macro trends and daily performance insights.
Building an MMM
Creating a model for MMM is a bit like being an artist and a scientist at the same time. You have to create variables out of various marketing efforts to factor them into the model - this could be how much you're spending on ads, how often you're running promotions, or how much you're changing your prices. Once the variables are created, multiple iterations are carried out to create a model which explains the sales trends well. This requires both systematic methods and a bit of human intuition.
Building a Marketing Mix Model (MMM) involves a crucial step of collecting and cleaning the data, and getting it in the right format for modeling. Each row in your spreadsheet or database corresponds to a day or week, and every column represents an independent variable in the model, with sales volume or value as the dependent variable. It relies heavily on a deep understanding of the business or industry and expertise in MMM methodologies to create variables that are both relevant and accurate.
A key discussion in the MMM world revolves around using modern automation versus what could be called an “artisanal” approach. Different practitioners land at different points on the spectrum based on their capabilities and philosophies. The variety of successful MMM-building methods reflects the diversity of marketers’ needs. Traditionally, MMMs were built manually by econometricians. But today, some argue automated tools can eliminate human bias and produce faster insights from huge datasets. Others insist that human judgment and expertise remain essential. Automation alone may produce spurious correlations or miss key drivers.
On the other hand, automated MMMs can quickly analyze millions of data points to uncover patterns that humans may miss. This enables frequent model updates to guide real-time decisions, rather than waiting a few months for a model update. Removing human bias from the model is of paramount importance, particularly in MMM where the modeler has a lot of leverage to make the model say what the client wants to hear.
MMM Project Timeline
The process of developing a Marketing Mix Model (MMM) can be broken down into the following steps, each with its own time frame, based on Meta’s Analysts guide to MMM:
- Define Your Destination: Questions and Scope (1-2 weeks)
First, determine what you want to achieve and the scope. Ask questions like: Do we want to optimize media mix? Measure promotion impact? Identify new customer acquisition drivers? Defining a clear purpose guides an MMM that answers your most pressing needs.
- Gather Provisions: Collecting and Reviewing Data (5-8 weeks)
Data collection and review are the most crucial—and time-consuming—parts of your journey. Work closely with analysts to gather sales, media, promotions, and other data. Review meticulously to ensure accuracy. Quality data inputs = quality insights. Plan ahead and anticipate delays to avoid frustration.
Navigate Carefully: Modeling the Data (4-8 weeks)
Modeling involves refining algorithms to detect relationships between your data points. It requires patience and trust in the process. Check-in with your analysts, but avoid reactionary changes. Multiple iterations are typical and necessary to achieve a balanced, insightful model.
The Destination and Beyond: Analysis and Recommendations (2-4 weeks)
Finally, you’ve arrived! Analyze the outputs and metrics to determine key insights and strategies, guided by your original questions and needs. But the MMM journey doesn’t end here—your model should be updated regularly to continue optimizing your marketing as needs and trends evolve.
The MMM path typically takes 3 to 6 months. But with the right mindset and preparation, the rewards of data-driven marketing insights and strategies are well worth the effort. As with any journey, there may be delays and obstacles—so focus on your destination, learn along the way, and trust your guides.
Adstocks and Diminishing Returns
As marketers, we often make simplistic assumptions about how our efforts impact results. We hope each ad directly leads to sales and that more marketing always means more business. The realities, however, are far more complex. Two key concepts, adstocks and diminishing returns, illustrate why advanced tools like MMM are critical to truly understanding marketing’s effects.
Adstocks: Delayed Impact
Adstocks refer to how an ad’s impact reverberates over time through our targets’ memory and word of mouth. Rather than disappearing after it airs, an ad’s influence fades gradually like a depleting stock. MMM uses adstock transformations, a type of weighted average, to capture how fast this “echo” decays to optimize your mix.
But determining an ad’s decay rate is challenging and requires testing different rates to find the best model fit. For example, you might find that video ads on YouTube have a 1-week half-life (the impact of a dollar spent today, continues to onwards, halving every week), whereas short-term channels like Google Search Ads don’t have any ongoing impact.
Diminishing Returns: Channel Saturation
The more you spend on a channel, the less you gain from each additional dollar—a diminishing marginal return. At some point, the returns plateau, and the channel is “saturated.” Determining this saturation point is key to allocating budget where it’s most impactful.
For instance, a bank I worked with showed diminishing returns beyond $20,000 per day on Facebook ad spend, which meant we had to find other channels to expand. Seeing where on the curve we were for each channel let us plan out exactly what was realistic in terms of our quarterly goals, and concentrate our efforts on the right areas.
Scenario Planning in MMM
MMM provides more than historical insights—it enables you to simulate future scenarios and prepare strategic responses. You can envision how the market may evolve and test how different mixes or spend levels may impact your results, gaining a competitive edge. This is how you translate insights into action.
Scenario planning helps you think several moves ahead, like a chess player anticipating the other’s moves. Each scenario represents a possible future state of the market, channels, customer segments, economic factors, or competitors. You define scenarios based on trends, then use your MMM to predict the outcomes of different marketing responses for each situation. You can even reverse-engineer what budget allocation maximizes your return across channels.
The Challenge of Correlation in MMM
MMM can struggle when there's too much correlation between factors or lack of variation in ad spend. This statistical phenomenon is called Multicollinearity, where two or more predictors in a model are highly correlated. For instance, if a TV ad and a radio ad always run at the same time, the model may struggle to distinguish which one is driving sales. This can result in unstable estimates in the model, as small changes in the data may lead to significant changes in recommendations, rendering them unreliable.
The challenge of accounting for long-term effects is another important aspect to consider. Marketing activities often have effects that last beyond the immediate period in which they are measured, as we highlighted with adstocks. For instance, a TV ad might lead to increased sales not just in the week it was aired, but years later (“Just Do It”, “Think Different”, “Got Milk”). These longer-term effects can be difficult to capture accurately in MMM with adstocks, but failing to account for them can lead to an underestimation of the effectiveness of marketing activities.
This is another great example of why forward-looking brands are combining MMM and incrementality - often using incremental lift tests to validate MMM results, or in more advanced cases, using incremental lift tests as “priors” in their MMM model. This helps ground their correlative MMM practice in causal-based measurement, as incrementality tests measure the true causal impact of media on sales.
Trusting the Model
An MMM is only as valuable as the trust you can place in it. While MMM provides key insights, its quality and limitations determine if you can rely on its strategic guidance. Follow these best practices to develop an MMM you can believe in:
A poorly built model can misattribute marketing impacts, leading to misguided decisions. To ensure quality:
- Review diagnostics to understand performance and limitations. Check metrics like R-squared, RMSE, and MAPE to gauge how well your model fits the data.
- Validate predictions against actuals. Compare the model’s forecasts to what truly happened to confirm its accuracy before using it prescriptively.
- Balance accuracy and practicality. An overly complex model may be highly accurate but uninterpretable. Keep your model accurate but also plausible and transparent.
- Conduct in-market tests. Make controlled changes to your marketing mix in select areas and compare results to the model’s predictions. Recalibrate as needed.
It’s the last one that is the most important because running experiments is really the only way you can truly establish a ‘ground truth’ for whether your model is accurate. Once you have the results of lift tests, for example, by switching off spend in 5 US states and comparing them to the other 45 to measure the uplift, you can use those results to improve the accuracy of your MMM.
Marketing Mix Modeling is a powerful tool that can help you make more informed decisions about your marketing strategy. Especially for brands that have years of data, it can provide analysis at a high level. You can understand which parts of your marketing mix are really driving your sales, simulate future scenarios, and adjust your tactics for optimal results.
But like any tool, it's not without its challenges and limitations.
When marketers need more granularity and ongoing insights at a tactical level, MMM comes up short - it can’t offer insights into campaign-level performance and decision-making.
However, incrementality provides the needed insights to tie media spend directly to business outcomes, providing clear allocation and optimization recommendations. To diminish correlation/causality issues, Measured recommends a combination of MMM and incrementality testing. Measured is interoperable with both 1st party (in-house) and 3rd party MMM providers, making it easy for brands to leverage incrementality testing and MMM together.
It's important to approach MMM with a critical mind, always questioning the assumptions behind the model and looking for ways to validate and improve it. The key is to make MMM just one of the tools in your arsenal, and comparing it to platform reports and the results of scientific lift tests is what gets you closer to the truth of what’s working.
Whether you're an ecommerce startup trying to figure out where to spend your ad budget, or a large corporation trying to understand the impact of your TV ads, MMM and incrementality testing should be a part of your measurement strategy.
Want to learn more? Schedule a demo today to see how Measured can help your brand grow.