The Granularity Trinity in Marketing Mix Models

  • The Need for Granularity and Actionability 
  • The Granularity Triangle/ Trinity in MMM 
  • The Challenges faced by Granular MMMs   

Introduction

Granularity and actionability are the holy grail of all marketers in relation to marketing measurement. This wasn’t accessible around 20 years ago, mainly because data wasn’t nearly as available as it is today, the methodologies used not nearly as advanced, and the computation power not nearly as fast. Today, the availability of these elements has paved the way for highly granular MMMs to be a reality. This has been further boosted by MMM specific data feeds that are abundantly provided by tech giants like Google and Facebook.

On the other hand, the proliferation of media channels, namely digital media, have made the optimization more challenging and opened the door for businesses to be more demanding. Naturally, MMM had to evolve from an outdated to a more contemporary version of itself. And it had to remain up-to-speed in relation to the business requirements and expectations. 

The Granularity Trinity in MMM 

Granularity depends on the type of business modeled and the precision of the business question that should be answered. In other terms, what is highly granular for a brand could be shallow for another. But regardless of the degree of granularity the business aims for, it will be done within the boundaries of the granularity trinity. We define the three dimensions of granularity as: 

  • Variables 
  • Regions  
  • Customer Segments  

These dimensions could be cumulative or mutually exclusive. And the use of a contemporary MMM methodology allows acknowledging and addressing this granularity triangle. This, in turn, will result into higher accuracy, robustness, actionability and more sensitivity to real consumer behavior. 

    1. Granularity at the variable level 

    ROI, contribution and campaign performance vary massively depending on the quality of the media used and the creative execution. Not all campaigns will have the same impact as execution. It is therefore crucial to contextualize the campaigns. Proper contextualization goes beyond the share of budget and extends to campaign execution elements. Campaign splits are now possible at different levels Because many online platforms made impressions and spend available at different data cuts, e.g. Ad Format, Campaign Objective, Ad Id, Placement. This ensures that different campaign execution features are assessed and measured as part of MMM. However, we can’t just measure the impact of media and campaigns as a bucket and ignore the differences at the execution level. Cutting data to the level it was executed at (e.g. platform, creative, the day part, the length, the format, or the delivery) is important to ensure accurate and actionable measurement.   

    Splitting variables is key for better measurement. It allows the analyst to determine if a particular element, creative for example, has a bigger impact than another. This will translate into adding more variables into the model. 

      2. Regions:  

      Regionality refers to splitting data to sub-national sets and building regional models. This type of granularity requires the analyst to pool the data and instead of looking at total sales for the whole territory, they can enhance granularity by modeling several regions. The reason here is the belief that certain regions might have different sensitivity to media. Thus, they need to be analyzed separately. This will help the business understand the customers’ behavior and reactions to their media, and in consequence design their content accordingly. It also maximizes variation and data points which consequently increases confidence in all measures and the likelihood of measuring an activity, especially when this happens at the regional level (e.g. regional outdoor activity),

      → What is measured through regionality is tricky to measure at the national level as it could be diluted by the effect of the rest of the regions, especially if the region in which the activity happened is small compared to National (total). 

      3. Customer segments 

      The behavior of a brand’s customers could vary significantly even within the same region. Creating a set of sub-segments will help the business isolate patterns of their customers’ behavior. As a result, they can understand how each segment reacts to their media, the contribution of each segment in their total sales, and which segments are more loyal to the brand. For example, in the banking sector, they cluster their segments based on age. And the message they disseminate varies based on the segments and their responses.  

      → When the clusters are not pre-set, the MMM partner needs to start by doing a cluster analysis exercise whereby customers are clustered based on behavior, then test their responsiveness to media.  

      If data is collected at a customer level, the regional level, and the granularity of the variable, one could get a comprehensive view of all the granularity that MMM can offer the business. They will be able to crack the code of what could be the best message that they should channel, the best platform and region to channel it, and the customer segment to whom the platform should be targeted. 

      → You could work for brands that are interested in looking at very granular cuts of the data for different definition of regions and different customer segments within the regions.  

      The main challenge of highly granular MMMs: 

      The lack of good quality data is the biggest hurdle faced by the measurement industry. There is a huge gap between what is expected of marketing analysts to deliver and the data with which they are provided. Even the biggest brands sometimes deliver data that suffers from anomalies and discrepancies. This inhibits the extraction of granular results, and lead to the disappointment of the brand.  

      This dilemma could not be solved even with the most state-of-the-art software.  Advanced technology will help if only there is a good data set. In that, it could create extra variables, and automate transformations that mimic consumer behavior, for example, the synergy effect, the carry over effect, diminishing returns, halo, etc. But technology cannot tell a story if its roots aren’t in the data set.  

      The analyst needs to set the business’s expectations about the granularity of their results depending on the data they’re provided and its quality.  

      Expert’s Opinion:

      A piece of advice, take baby steps. Your first iteration does not need to be elaborate and granular. Start with a simple ROI percentage contribution, then drill further in the following iteration and tackle regions, then customer segments…  

      Conclusion:

      All this explains the movement to MMM after two years of cookie-based methods. There is an awareness that what was provided was not magic and certainly was not exclusive to cookies and that MMM is able to provide the same results without breaching customers’ privacy; even the biggest players in the digital space know that brands are savvy for granular results, and they are providing the granular data needed 

      MMMs are adjusting fast to this change and the granularity trinity/ triangle is the successful formula they are using in its adaptation. Not only does this formula work because of how adaptive and efficient it is but because it is in sync with the privacy regulation in place. Thanks to the fact that MMMs can reach deep levels of granularity without divulging the customers’ identities, it is safe to say that they have claimed their space in the current map of Measurement.