Understanding the Role of Priors in Media Mix Modeling (MMM): Webinar Recap

Last week, Rockerbox hosted an insightful webinar diving deep into the role of priors in Media Mix Modeling (MMM). Our expert panelists, Eddie Chou (Product Manager) and Mitchell Jones (Data Scientist), walked us through the intricacies of priors, their influence on MMM, and best practices for leveraging them effectively. If you missed it, here’s a recap of the key takeaways.

Watch the session here.

What Are Priors in MMM?

Mitchell kicked off the discussion by defining priors in simple terms: Priors are the foundational beliefs or assumptions used in MMM to guide the model in understanding how media spend interacts with key performance indicators (KPIs). Since MMM is a simulation-based process, priors serve as the initial parameters that shape how the model explores potential relationships between media spend and business outcomes.

For example, if you believe a particular media channel has an ROI of 3, that belief—along with a range of other potential values—gets incorporated into the model. The MMM then combines this prior belief with actual data to refine and generate the final results.

Why Are Priors Important?

Priors play a crucial role in MMM for several reasons:

They provide a starting point for exploration. Without priors, the model would have no directional guidance, leading to potentially unrealistic or erratic results.
They balance evidence with expectations. By incorporating priors, marketers ensure that their models are grounded in reality, while still allowing data to influence outcomes.
They improve model stability and believability. If MMM produces results that contradict all known performance metrics, it becomes difficult to trust and act on the findings.

Sources of Priors in MMM

Eddie broke down the four primary sources of priors that Rockerbox incorporates into MMM models:

Platform-Reported Numbers: Metrics pulled directly from ad platforms like Google, Meta, and TikTok. While helpful, these numbers often represent a best-case scenario, since platforms tend to “grade their own homework.”
Incrementality Tests: Gold-standard controlled experiments designed to measure the true impact of media spend on business outcomes. These are the most reliable priors and provide strong benchmarks.
Industry and Domain Knowledge: Broader marketing benchmarks that help set realistic boundaries for expected performance.
Customer Contextual Beliefs: Insights from marketers based on their experience and historical data, even if not formally tested.

How Priors Improve Model Performance

A key takeaway from the discussion was how priors enhance MMM’s effectiveness:

They provide guardrails to prevent unrealistic results. If an MMM without priors suggests an ROI of 20x for a given channel, that’s likely a signal that something is off.
They help resolve multicollinearity challenges. When multiple channels have overlapping effects, setting strong priors for some channels can help differentiate their individual impact.
They drive better decision-making. A well-calibrated model allows marketers to make informed budget allocation choices based on more reliable insights.

Balancing Priors for Optimal Results

Mitchell emphasized the importance of setting priors carefully to strike a balance between specificity and neutrality.

Overly strong priors can overly constrain the model, leading to results that simply confirm existing beliefs rather than uncovering new insights.
Too weak or neutral priors might allow the model to explore too many possibilities, increasing the risk of unreliable findings.
Iterative adjustments help refine priors over time, allowing marketers to test different scenarios and optimize their MMM models accordingly.

The Future of Priors in MMM

As MMM continues to evolve, the use of priors is becoming more sophisticated. Some trends to watch include:

More granular industry benchmarks: As Rockerbox gathers more customer data, priors can become increasingly specific to verticals and business models.
Greater integration of testing data: Rockerbox’s own incrementality testing solutions will provide more precise priors for modeling.
Iterative learning from MMM outputs: Ensuring continuity from one MMM refresh to the next by carrying over learnings to prevent drastic fluctuations.
Expanding priors beyond ROI: Incorporating priors for seasonality, ad stock effects, and diminishing returns to provide a fuller picture of media effectiveness.

Final Thoughts

This webinar provided valuable insights into how priors influence MMM and how marketers can optimize their use for better decision-making. If you’re currently using or considering MMM, remember that priors should be treated as data-informed professional hypotheses rather than rigid constraints. They should guide the model, not dictate it.

Want to learn more about Rockerbox MMM? Get a demo today.