Analytics Blog

Digital Data and the Future of Marketing Mix Modeling

Over the past several decades, Marketing Mix Modeling has been an important tool to assist firms in optimizing the allocation of budget to various types of media, such as television, radio, print, outdoor, and digital. In recent years, the digital component has grown out of proportion to more traditional media channels.

The growth of digital marketing has been propelled forward by the explosion of social networks, online forums, Wikis, product review sites, photo and video sharing sites, online news sites, streaming TV, and gaming. As a result, digital advertising and the use of influential messaging has grown, and marketers are considering allocating new levels of spending on digital media.

In order to implement digital marketing, methods have been explored and developed to link advertising exposure to consumer response. The advertising company tries to send the right ad to the right person at the right time. This is accomplished by first linking together devices, including smartphones, tablets, gaming devices, Roku sticks, Internet of Things (appliances, cars, etc.), and Home Automation devices (such as Amazon Echo, Google Home), and then delivering targeted content to the consumer who uses them.

With these newer methods of linking ad exposure to consumer response, ads can be rolled out and consumer response can be observed rapidly, using big-data platforms that feed dashboard engines.

In the face of these strong and disruptive trends, marketers may wonder, “Is traditional marketing mix modeling dead?” By “traditional,” I mean methods that use time-series, cross-sectional data with some form of econometric modeling to measure and predict ROI.

The traditional modeling approach, compared to the use of digital technologies to roll out and measure marketing ROI:

• Is slow to execute.
• Cannot project ROI for levels of media spend far beyond that which exists in the data.

This latter limitation hit me square in the face recently when a client asked me, “Can your marketing mix model tell me the ROI of tripling my digital spend?” My gut response was, “Yes, we can do that, but we have to make some assumptions about increasing or decreasing returns to the incremental dollars spent.” Some have addressed this limitation by applying agent based models. While this approach is more sophisticated and can project beyond historical levels of spend, it frequently uses assumptions about ROI embedded in equations that explore “what if” scenarios. In other words, no real measurement of ROI is used.

Still, traditional marketing mix modeling boasts advantages as well:

• It uses solid econometric modeling methods that minimize measurement bias.
• It provides the foundation of optimization tools that assist marketers in determining how much budget to allocate to various media categories, both digital and nondigital.

For these, reasons, traditional marketing mix modeling is not dead, but it must be transformed by incorporating learnings from real-time, digital advertising experiments in a mathematically consistent fashion.

For example, suppose a particular marketing mix model, due to the absence of historical data for high levels of digital spend, cannot answer the question, “What would happen if I were to triple my digital spend vs. my historical spend levels?”

To answer this question, one could design an experiment (to be executed in the digital domain) to measure the incremental ROI of digital spend with separate test cells for different levels of spend. The levels of spend could extend from below the maximum historical spend to double or triple the maximum. The results from such a digital experiment could then be used in a mathematical equation to extrapolate the marketing mix model ROI of digital spend to levels of up to triple the historical maximum spend.

While a marketing mix model might take several months to develop, once developed it would provide the framework to rapidly incorporate results from such real-time digital experiments. The enhanced marketing mix model would still be used to optimize media budget, but it would no longer suffer from its chief limitation, namely, the inability to project media ROI beyond the range of available historical data.

John Colias, Ph.D.

Senior VP Research & Development

As a leader with both university teaching and business consulting experience, John focuses on predictive modeling, prescriptive analytics, and artificial intelligence. As Senior Vice President, Research & Development, at Decision Analyst, John combines academic and business interests to help analytics professionals by offering cutting-edge analytic solutions tempered by business realism. He holds a doctorate in economics from The University of Texas at Austin, with specializations in econometrics and mathematical modeling methods.

• 
•