What’s Media Mix Modeling? [Marketer’s Guide with Examples]
Have you ever felt in the dark when it comes to understanding the real impact of your marketing dollars?
Determining where and how conversions are occurring is crucial in optimizing your budget to drive the most impact with your marketing budget. But in a fragmented media environment, optimizing your marketing performance is challenging. How can we measure how our paid TikTok ads impact instore sales? We know our target audience watches a certain television channel, but during which time slots? Are those TV ads even providing a positive return on ad spend?
Media mix modeling (MMM) holds the answers to all of these questions and more. It’s a powerful tool for marketers seeking to optimize their marketing investments. By providing a holistic view of how various factors contribute to sales and conversions, MMM enables data-driven decisions that enhance marketing efficiency and business growth.
In this article, we explore how media mix modeling works, and will provide examples of how MMM drives smarter ad spend decisions.
What Is Media Mix Modeling?
Media mix modeling (MMM) is a statistical analysis technique used to measure the impact of marketing activities across multiple channels. It accomplishes this by ingesting historical data, then running a regression analysis to understand the relationship that independent variables (like channel marketing spend) have on dependent variables (such as net new sales).The process isn’t one-and-done either – the model should continuously optimize itself based on actual consumer responses and business outcomes. With each iteration, marketers are able to gain an even higher degree of clarity and certainty in their everyday decisionmaking.
The method is similar to marketing mix modeling, although media mix modeling focuses on optimizing the performance of promotion channels, rather than other aspects of the marketing mix (like personnel, place of sale, and pricing).
To put it simply, MMM uses the predictive and causal capabilities of regression analyses to accurately explain how a combination of disparate marketing activities lead to a single desired effect. For most marketers, this desired effect is improved ROI – but MMM can also be used for more complex purposes, like measuring brand equity or predicting stockouts. Here are a few specific examples:
- Which of our marketing efforts are having the biggest impact on reaching our goals?
- How does seasonality impact the ROI of geotargeted display ads?
- Are our brand building efforts actually helping us improve loyalty, or are we just wasting spend?
- Will more press coverage help us meet our share of voice goals in the next six months?
- Are any audiences purchasing more or less frequently due to emerging market trends?
- Do we make more money if we send out a promotional email with a 10% off coupon or with a free gift?
- How should I allocate marketing spend across channels to optimize short-term return on ad spend?
- If I optimize for short-term returns, how will that spend allocation impact long-term marketing performance for other channels?
Marketers can answer these questions and determine the true influence of media buys across diverse platforms by using a simple byproduct of their marketing campaigns: Aggregate data. Brands can set up data feeds for online and offline advertising channels like social media, print advertising and streaming TV advertising. Then, the data is layered with other performance-related factors, seasonality and economic conditions. From there, marketers can use that data to understand how each touchpoint and externality impacts outcomes like sales revenue, brand health, new customers, and conversions.
Media mix modeling is a top-down privacy resilient approach to attribution in the sense that it applies statistical methods to analyze and interpret marketing data, providing a systematic understanding of how different marketing channels contribute to overall business goals in the broader context of the market. The quality of insights derived from MMM heavily depends on the quality and granularity of the data used.
— Annica Nesty, Group Director of Marketing Science at Tinuiti
Behind the Model: How Media Mix Modeling Works
In our modern world, we’re surrounded by all types of algorithms and statistical models. While we use these processes almost every day, it’s not always apparent how they work. However, with adoption on the rise, marketers should have a basic understanding of how media mix models work behind the scenes.
The Media Mix
First, the concept of the “media mix” emerged in the 1950s from Neil Borden, a Harvard professor of advertising. He thought of marketers like chefs in a kitchen, trying to find the perfect mix of ingredients to induce a consumer’s response. As the decades went on, researchers tried to define exactly what that perfect mix of ingredients were.
Since each business, product, and consumer is unique, 20th century marketers probably felt they were searching for the holy grail. “The perfect mix” was, inherently, a moving target.
Innovations in Statistics
Speaking of moving targets, in the early 1700s someone named Isaac Newton was trying to predict how the movement of the Earth impacts celestial alignments – specifically, the equinox. Since the earth is tilted, the equator doesn’t easily align with its orbit of the sun. There were independent variables (such as the earth’s axial tilt, gravitational influences, and the time of year) that produced a dependent variable: The exact time of the equinox. While Newton didn’t quite crack the code, he came close to developing the first statistical regression model. It would take 200 more years for mathematicians to develop a reliable model for predicting the earth’s position in the cosmos.
Unfortunately, running regression analyses was extremely laborious in the early 1900s. Accurately calculating regressions could take a day or longer of punching numbers into simple digital calculators – and with computers being few-and-far-between until the 1980s, it wasn’t feasible or obvious to use regression to find the perfect marketing mix.
Synthesis
While large companies like Coca-Cola were using basic regression-based MMM as early as the 1970s, it was exceedingly uncommon to see marketers leverage media mix models until the 1990s. This is due to two key advancements: First, computers weren’t just becoming more powerful – they were becoming more affordable and ubiquitous as well. Second, statisticians were using computers to research new regression models like Bayesian learning and time-series. This allowed models to place less importance on historical data and outputs, while prioritizing more timely and relevant information. Now, MMMs wouldn’t be burdened by historical data – rather, it could be used to massively improve accuracy over time.
However, it wasn’t until the 2000s that the third key advancement brought a spark to the powder keg: The rise of digital advertising. Compared to traditional advertising, internet advertisements allowed brands to access granular and timely ad performance data. This strengthened the reliability of MMM models. On the flip side, the internet also brought a lot of uncertainty to the world of marketing attribution. This enhanced the demand for MMM, which promised clarity.
More than twenty years later, MMM is still iterating and improving, especially with the rise of artificial intelligence and deep learning. At this point, media mix modeling is an indispensable tool for measuring the impact of media spend and predicting the future outcome of marketing investments.
Media Mix Modeling Examples & Case Studies
With the right media mix model, a business can measure their past marketing performance to improve future ROI by optimizing the allocation of the media budget by channel and/or tactic, including: traditional and digital media channels, promotions, pricing, competitor spend, economic conditions, weather, and more.
1. Linking digital media to in-store sales
Poppi, a top-selling prebiotic soda brand, wanted to understand how their digital campaigns influenced in-store purchases. However, the problem was that they utilized a lot of promotion channels, and sold their drinks across thousands of retail locations. While they had clear results for digital sales from platforms like Amazon Online Video, Instacart, and TikTok, measuring the impact on in-store purchases was a challenge.
To address this, our team worked with Poppi to leverage Crisp, a platform that connects real-time sales data from retailers down to individual SKUs and ZIP codes. Then, our software experts developed a custom model in Bliss Point that analyzed the relationship between Poppi’s digital campaigns and in-store sales across different geographic regions.
The model gave Poppi a detailed breakdown of how each campaign influenced in-store purchases, allowing them to optimize their media mix for maximum ROI. One key insight was that TikTok ads made their target customer 80% more likely to purchase Poppi instore, leading them to allocate more of their budget to this platform.
2. Forecasting revenue and maximizing ROI
An international ecommerce brand wanted to forecast their second-half of the year and create an optimal media mix to make their marketing dollars work smarter. A combination of the client’s data, marketing data, and machine learning were required to create a powerful, custom media mix model.
To build the model, the business used over 2 years of digital marketing and revenue data, analyzing it by market, tactic, and day. The data was then used to create model to assess future spend showing how changes in investment across channels could impact revenue and sales.
The full digital media mix model gave the ecommerce brand a detailed analysis of where to optimize their spend across all digital marketing channels.
One recommendation was to shift dollars away from social—which historically had been at or near 30%—to paid search. This recommendation came with another layer of insight: The brand realized they were overinvesting in awareness campaigns, and needed to invest more heavily in capturing current demand during the second half of the year.
Working with a robust media mix model, the brand was able to break down how much media spend was needed by each channel in order to achieve the 30% YoY revenue goal they targeted.
Geo-based incrementality testing
A premium luggage brand wanted to test the impact of YouTube ads on driving incremental site visits and purchases during the holiday season. Specifically, the brand wanted to measure the effectiveness of its always-on spend and audience-targeting efforts on YouTube.
To execute this test, we worked with the brand to leverage Google’s external geo-based incrementality testing feature, integrated into their Bliss Point platform. This feature allowed our teams to measure not just site visits and purchases, but also the incremental impact of these conversions. Using the brand’s own historical data to target the right audience with tailored ads, they ran the test during the 6-week holiday season and reported back with actionable insights on how YouTube ads drove incremental results.
This test was highly successful. It drove a 7% incremental lift in site visits, representing 202,000 additional site visits. This turned into a 6% incremental lift in purchases, clearly demonstrating the value of external geo-based incrementality testing on YouTube for driving retail performance.
The Benefits & Challenges of Media Mix Modeling
MMM helps you accurately connect all the dots, leveraging (ideally) a wealth of provided data, to understand how disparate aspects of marketing campaigns work together in helping you reach your business goals. The nature of these benefits is multifaceted, offering marketers a strategic edge in navigating the intricacies of their advertising efforts. Let’s dive into each benefit in detail:
- Omnichannel Campaigns: MMM excels in providing insights for omnichannel campaigns, allowing marketers to understand and optimize the impact of their initiatives across various channels. This capability is crucial in today’s interconnected digital landscape, where consumers engage with brands through diverse platforms.
- Improved Oversight Over Media Spend Impact: MMM provides a comprehensive view of the impact of media spend, enabling marketers to assess the effectiveness of their investments. This improved oversight ensures a clearer understanding of how each component of the media mix contributes to overall campaign success.
- Media Spend Optimization: With MMM, marketers can optimize their media spend by identifying the most impactful channels and touchpoints. This data-driven approach allows for strategic adjustments in budget allocation, ensuring that resources are directed towards the avenues that yield the highest returns.
- Effective Targeting of Audiences: MMM’s analysis helps in refining audience targeting strategies. By understanding which elements of the marketing mix resonate most with specific demographics, marketers can tailor their campaigns to effectively reach and engage their target audience segments.
- Forecasting with Certainty: One of MMM’s strengths lies in its ability to forecast results with a high degree of certainty. This forecasting capability empowers marketers to make informed decisions based on predictive analytics, aiding in long-term planning and goal setting.
- Reduced Reliance on Personally Identifiable Information (PII): MMM minimizes the reliance on personally identifiable information for analysis. This is especially crucial in an era where privacy concerns are more important than ever.
In the post-cookie and post-IDFA landscape, where privacy concerns and regulatory changes limit access to individual user-level data, media mix modeling has become a pivotal analytical tool. MMM’s emphasis on overall marketing spend allocation and its proficiency in establishing cause-and-effect models, address the challenges posed by the diminishing availability of explicit conversion information, providing marketers with a privacy-respecting and insightful approach to navigate the evolving digital advertising ecosystem.
Overcoming the Challenges of Media Mix Modeling
Overcoming the challenges of media mix modeling (MMM) involves addressing key factors like model fit, data quality, and the frequency of insights. Let’s take a closer look.
1. Ensuring the model is a good fit
A strong media mix model should be accurate, adaptable, and grounded in reality. This can be a challenge, because MMM makes its predictions based on historical data – meaning that if a model isn’t built with care, it will just display results that are too dependent on your previous data.
For example, brands looking into MMM should ensure the model has measures to prevent overfitting. Overfitting happens when a model clings too tightly to historical data, basically showing what’s already happened in a similar situation, rather than making new insights based on real market trends. It might seem highly accurate in theory, but will struggle to predict future results. Great MMM models should use cross-validation and base their predictions on a wide variety of internal and external factors.
The importance of accounting for external influences can’t be overstated. For example, imagine that you’re running advertisements on the Boston subway. You’ve been consistently building brand value in the metro area, until ad performance suddenly drops in November. A basic MMM wouldn’t explain much – just that your subway ads aren’t performing well in this market.
This very small lack of clarity can lead to massive mistakes in decisionmaking. Intuition might tell us this is caused by seasonal ridership declines, leading you to reallocate spend to billboards on the highway. But in reality, the true problem could be advertising fatigue caused by oversaturating the market. In this case, your simple assumption could lead to wasted spend and damaged brand equity.
2. Data volume & data quality
Working with high-quality data is important in any measurement initiative, but for MMM to work effectively, it also needs a lot of data to build a reliable model. Remember, the model is analyzing complicated relationships between multiple channels, touchpoints, externalities, and more, to determine a business outcome with a high degree of accuracy. The high volume of data helps it detect patterns, while the quality of that data ensures those patterns aren’t spurious correlations.
For example, if you want your model to analyze the impact of seasonality on a planned media buy, it ideally needs at least three full seasons (or three years) of data to recognize patterns and produce accurate insights. However, this doesn’t mean you need to wait three years after implementing an MMM solution to start seeing seasonality-driven insights. Instead, brands can connect an MMM platform to historical data sources from digital media platforms they have already been using for years, allowing the model to leverage existing information from day one.
One way to meet this demand is through automated, programmatic data connections, such as integrating data from Facebook’s API or streaming TV providers like Paramount. These connections feed the model with actual campaign data from all corners of your marketing strategy inputs, providing a rich and trusted source of data.
3. Frequency of readouts
Generally speaking, MMM is a ‘long game’ initiative with infrequent reporting by its nature. Brands and advertisers who are more accustomed to daily or weekly updates may struggle with ‘waiting out’ the analysis. This is because MMM uses aggregate data – not user-level data – to produce its performance insights. For that reason, most of the basic models on the market offer limited insights on brand impact, personalized targeting, and customer experience.
While most of the standard models on the marketplace can only offer monthly readouts and require months of training to return reliable information, this challenge is becoming less pronounced due to recent advancements in MMM. Models like Bliss Point by Tinuiti use “Rapid Media Mix Modeling” (rMMM) to provide highly granular insights with speed, precision, and transparency. While these models are few and far between, it is possible. So, make sure your chosen MMM vendor is able to provide a high quality of information as frequently as you need it before you sign any contracts.
Common Misconceptions About Media Mix Modeling
Media mix modeling, like many other analytics solutions, has also become a marketing buzzword that has generated its fair share of misconceptions.
Here are a few of the most common misconceptions around media mix modeling.
Media Mix Models Are Not Transparent
With large datasets and statistical analysis involved in media mix modeling, the methods behind the technique have been critiqued for their obscurity. If there is no perceived transparency in the process, how does a brand know if its media mix model is really accurate?
Any organization specializing in media mix modeling should provide a transparent approach, with deliverables such as outlines, milestones, and performance reports. Additionally, you may want to consider partnering with an agency that truly understands how media mix modeling aligns with your needs and expectations. Every business is unique and each media mix model is based on multiple factors.
Media Mix Models Do Not Provide Real-time Data
Today, results are often measured by the timeliness of their delivery, with the current digital marketplace allowing for almost instantaneous real-time data. Media mix models do actually provide compelling real-time marketing insights, perfect for evaluating new campaigns, new competitors, and assessing pricing actions or changes in promotional strategies.
A powerful partner in media mix modeling will provide sophisticated tools and real-time approaches to satisfy your business performance assessments. Your partner should also be able to provide forecasting, simulation, or AI- and machine-learning-integrated models to suggest future movements.
Media Mix Modeling is Biased to Offline Channels
Though media mix strategies do integrate and consider offline channels in their approaches, media mix modeling also considers all digital channels — including display, email, paid search, social, and more. Remember—it’s considering your media mix. If that includes ten different channels and you provide enough high-quality data for each, they will all be considered in your marketing mix analysis.
In fact, as customers have become more intertwined with digital channels, media marketing models have adapted to go even deeper into the analyses provided by those channels’ respective insights to support better budgeting choices and customer segmentation reports.
Media Mix Modeling vs. Data-Driven Attribution Modeling
Like media mix modeling, attribution modeling also studies the efficiency of marketing strategies — but there are important differences.
Attribution modeling is a general term that refers to tracking engagement to better understand how specific tactics drive action at the user level. This modeling works well for analyzing specific customer touchpoints, focusing on elements like how a consumer converted, which creative on which channel led to that conversion, and what the expected ROI could be if more ad budget were shifted to that channel.
Media mix modeling takes a higher-level, more comprehensive picture. This modeling isn’t designed to measure user-level engagement like impressions and clicks, rather its primary function is measuring the impact of an entire touchpoint on specific marketing objectives.
Data-driven attribution modeling and MMM each have their own set of strengths. It’s not a matter of one being better than the other, rather one being better-suited to different types of marketing analysis.
For example:
- The precision of the data-driven attribution: Let’s assume you want to invest more spend in a social ad campaign during the holiday season. While MMM is an option for determining where to allocate those dollars, data-driven attribution excels in dissecting the intricate customer journey, offering a microscopic view of user interactions. For instance, if you’re keen on understanding the exact value of a single click from your social media campaign, data-driven attribution can illuminate the path.
- The holistic perspective of the media mix modeling: Media mix modeling, can consider the impact of offline actions and initiatives. Unlike the more narrowly focused attribution models, which might overemphasize the first or last touchpoint, MMM assesses the collective impact of all channels over time. This makes it an indispensable tool for strategic planning and long-term investment decisions in your marketing portfolio.
“Attribution modeling is based on a bottom-up approach while media mix modeling takes a top-down approach. Media mix modeling provides a long-term view of the marketing ROI of media activity, while attribution modeling evaluates individual-level activity to provide a short term view of marketing ROI.”
— Annica Nesty, Group Director of Marketing Science at Tinuiti
Conclusion: MMM Closes the Loop on Marketing Performance
In an ever-evolving digital landscape, MMM’s adaptability to the post-cookie/post-IDFA world positions it as an essential tool for marketers. As businesses seek to connect the dots, leverage data, and make strategic decisions, MMM is a crucial ally in the dynamic realm of mixed media advertising.
At Tinuiti, we know, embrace, and utilize MMM. Our Rapid Media Mix Modeling sets a new standard in the market with its exceptional speed, precision, and transparency.
Our proprietary measurement technology, Bliss Point by Tinuiti, allows us to measure what marketers have previously struggled to measure – the optimal level of investment to maximize impact and efficiency. But this measurement is not just to go back and validate that we’ve done the right things. This measurement is real-time informing what needs to happen next.
Curious about how we can tailor strategies to hit your unique marketing bliss point, including Rapid Media Mix Modeling? We’re eager to chat. Contact us today for details.
You Might Be Interested In
