Back to the Future: Trends and Innovations in MMM

In the evolving landscape of marketing measurement, Marketing Mix Modeling (MMM) is witnessing significant advancements. Here, we explore three key trends shaping the future of MMM: the emphasis on causality, the adoption of Bayesian methods, and the push towards real-time analysis.

Emphasis on Causality

The advertising domain is filled with correlations. It’s common to observe patterns, such as increased social media spending coinciding with higher sales. However, these correlations do not necessarily indicate a causal relationship. Understanding the causal impact of each marketing dollar is crucial for making actionable decisions and optimizing budget allocation. For example, branded search might show a strong correlation with revenue due to increased demand, not because additional spending on branded search directly increases revenue.

Uncovering true causation within MMM requires a sophisticated approach. Simply fitting a regression model won’t reveal causal relationships. It involves careful consideration of which variables to include in the model, which to exclude, and how to calibrate the model with test results and domain knowledge.

The shift towards a causal inference framework is first a mindset change. It alters how you see the problem, what assumptions you make, and how you define success. Then, you need to use causal inference methods. In MMM, this is often done using calibration with geo tests and incorporating domain knowledge. These methods calculate the causal effects of specific channels and times. Incorporating these effects into your model means you are using ground truth for causal effect, forcing your model to fit based on this information and not solely on data patterns.

Google’s new open-source MMM model, Meridian, exemplifies this shift by emphasizing causality in its documentation and application. Meridian includes a dedicated section on “Causal estimands and estimation,” using causal inference terms and techniques to guide marketers in understanding true causal relationships.

The Rise of Bayesian Methods

Bayesian models are becoming the preferred choice for new marketing measurement tools, both paid and free, such as Google’s new model, Meridian. These models excel at incorporating prior knowledge and business assumptions, making them ideal for navigating the complex landscapes of modern marketing.

Bayesian models essentially create multiple simulations of possible realities based on prior knowledge, such as test results and domain expertise. For each simulation, they calculate the probability that the observed dataset originated from it, selecting the most probable scenarios. This approach inherently incorporates prior knowledge into the process. When new data about budget and sales is included, the model integrates this information with the prior distribution to form a posterior distribution, reflecting updated insights.

By incorporating uncertainty and prior information, Bayesian models provide a robust framework, and improve their causal reliability. In contrast, frequentist methods, like fitting a regression model, often yield wide or unreasonable results because they don’t incorporate prior knowledge as effectively. While these methods try to manage limitations manually, Bayesian models inherently incorporate these constraints, resulting in more robust outputs.

On the other hand, relying on prior data means you’re only as good as the data you rely on. A common issue is that test results often have a limited time frame. For example, if you conduct a geo test, measure the results for two weeks, and include that in your model, your Marketing Mix Modeling (MMM) takes this effect as the whole effect of your test. In reality, the impact of your test can continue for months due to ad hoc effects. Consequently, these tests often underestimate the true effect of the ads. By incorporating these short-term results into your model, you can introduce bias.

For instance, Meta’s Robyn model, a well-known frequentist approach, has been around for a few years. The creator of Robyn cited computational efficiency as the reason for choosing this approach. However, another advantage is that, since the test results are incorporated using a time frame, it handles the ad hoc problem better than most Bayesian models out there.

The adoption of Bayesian models in marketing measurement signifies a move towards more sophisticated and intuitive methodologies. By leveraging prior knowledge and updating with new data, Bayesian models provide marketers with powerful tools that ensure accurate and reliable insights. This trend underscores the future of MMM, where Bayesian models will play a central role in shaping marketing strategies.

Real-Time Analysis and Adaptation

One of the most exciting advancements in MMM is the push towards real-time analysis. Traditionally, MMM has been a retrospective exercise, relying on historical data to inform future decisions. However, the dynamic nature of today’s marketing environment demands more agile approaches.

Modern MMM solutions are increasingly designed to update and refresh often, incorporating new data as it becomes available. This real-time capability allows marketers to make adjustments on the fly, optimizing their strategies based on the latest insights. The goal is to create systems that not only analyze past performance but also provide immediate recommendations for future actions.

By integrating real-time data, marketers can stay ahead of trends and respond swiftly to changes in consumer behavior. This proactive approach ensures that marketing efforts are always aligned with current market conditions, maximizing the effectiveness of every campaign.

Conclusion

The future of Marketing Mix Modeling is bright, driven by an emphasis on causality, the adoption of Bayesian methods, and the move towards real-time analysis. These trends are transforming MMM from a static, retrospective tool into a dynamic, forward-looking system that empowers marketers with actionable insights. As these innovations continue to evolve, they will redefine how we measure and optimize marketing performance, ensuring that every marketing dollar is spent wisely.

By embracing these trends, marketers can stay ahead of the curve, leveraging advanced methodologies to drive better results and achieve their business goals. The future of marketing measurement is here, and it’s more exciting than ever.