Five Questions Practitioners Should Ask Before They Begin Their MMM Journey

Being a part of the digital world for the past nine years, I have witnessed our digital ecosystems constantly evolving. At present, data protection laws and anti-tracking policies are the epicenters of the digital universe. However, the industry has been able to pivot and find innovative ways to combat such situations. One solution to manage the current state would be moving away from consumer-level tracking to aggregated measurement studies. One such measurement solution is the marketing mix modeling (MMM); it is a time-tested statistical analysis thriving on historical sales data, marketing and non-marketing activities, and other variables that can impact an advertiser's business performance. It enables marketers to conduct future planning scenarios and better allocate resources between different channels and marketing tactics.

Advertisers commission MMM via third-party measurement partners, or the models can also be developed in-house if the advertiser has the right skills and tools incorporated within. It is essential for every practitioner beginning the MMM journey to ask themselves these five questions:

1. What internal data & analytics do I have available?
   • Data. The first and most important aspect would be to check internally on the availability of data. MMMs are purely dependent on quantity as well as the quality of data available. MMMs are not always about the media side of things but include a wide variety of data across the business to be able to run (e.g. consumer research, distribution, financial reports, competition, other market factors to name a few). MMM is the central holistic measurement study that brings together all these elements that interact together to either enhance or undermine the performance of the business.
   • Analytics. Practitioners can begin by checking if the advertiser has commissioned MMMs in the past via a third party. On the other hand, if the advertiser wants to move it in-house now and/or refresh methodologies, a cost- and time-effective method would be to look for any open-source MMM codes available. Facebook has recently launched an open-source project called Robyn intending to build a community of MMM methodologists, discussing innovation and contribution to open-source code to create automated MMMs using machine learning techniques.
      
2. Have I considered all the variables that can impact my advertiser's business? MMMs are a mix of art and science. The art is to make sure that practitioners have brainstormed with the stakeholders involved on all the variables that can potentially influence the business. I'd acknowledge that this is a bottleneck in the process and can take a long time, especially the first time around; but it's absolutely critical. Advertisers could/should proactively have data organized in order and have a data strategy, including quality assurance and sources & definitions of each data stream. The models can be misleading or drive inaccurate results if key variables are omitted during the modeling phase. Modeling should not be undertaken hastily or mechanically and should cater to the nature of every business to some extent.
    
3. Is the information collected granular enough to provide actionable insights? Granularity (in simple terms, level of information) is vital for modeling to offer essential insights and maximize model accuracy. A practitioner should ensure that they can provide the most profound insights for every channel and break down insights at campaign, product, and region levels. Facebook has its own dedicated MMM feed that breaks down the data daily. Granularity enhances the quality of information and shows every channel's true value. For more information, check out how granularity can unlock the full potential of MMMs.
    
4. How do I remove bias from my MMM models? Practitioners can look for biases in two ways:
   • Look at various statistical metrics to assess the model's accuracy; look for any multicollinearity between the variables; examine the R-squared and adjusted R-squared, MAPE, and Durbin Watson. Don’t forget to look at the predictive power of the model: check the out-of-sample forecast quality on the validation data set, as well as associated forecast errors and confidence intervals.
   • Calibration of MMMs with lift experiments run on and off Facebook can be critical to validate the MMM models. One of the validation methods is to compare the incremental value showcased in the model (incremental cost per KPI) with the platform's experimental tests. H&M MENA and Deloitte partnered together to run MMM and also calibrated results with lift studies outlined here.
      
5.  How can I use MMM results as valuable input for decision-making? MMM is a powerful measurement tool that provides both strategic and tactical insights for advertisers. It enables advertisers to understand the optimal cross-media budget allocation via future-facing scenario planners and saturation curves for strategic insights. For tactical insights, MMMs can deep-dive into individual channels and fuel optimizations by using granular information in the models.

MMMs are here to stay. Advertisers should invest in modernizing their MMM models, making them quicker and automated to understand the overall business performance from a bird's-eye view and dive deeper to derive actionable insights. The CMO needs to understand the potential of MMM and demand more. Practitioners need to work with other stakeholders and answer the right questions.