Although there are over twenty years of history in industry and an even longer academic tradition, there is still a relatively small community of individuals who have a solid understanding of marketing mix modeling. In an effort to help clients make better decisions in reviewing vendors and in acting upon recommendations, here are a few questions every CMO should feel comfortable in asking – or in entrusting your marketing research director to ask – your marketing mix modeler.
How are marketing effects being modeled?
We no longer live in a marketplace where mix models can only consider the direct impacts of marketing activities on sales. Today’s offline advertisements are generating traffic online, where consumers are exposed to paid search and display ads. This online behavior may be in the form of consumers’ searching for more information about the product or talking to their friends about the ad or product through social media outlets. How is your modeler or vendor accounting for the role of search and consumer-generated media in their mix models? Are they considering both direct and indirect impacts of marketing on consumer behavior? Y, that is to say consumer response, can no longer be considered a function of all the X variables at the same time!
How is the time dimension being treated?
Although every marketing mix analysis contains a time series of data, this is the dimension most often overlooked. Marketing does not occur in a static environment, yet most marketing mix modelers treat it as if it does. Customers exhibit feedback in their behaviors, whether out of habit or brand loyalty. Marketers behave dynamically by anticipating and/or responding to competitive or customer behaviors. Marketing drivers are even related over time in the sense that offline marketing activities drive consumers online to search, where they are exposed to paid search advertisements.
Most modelers consider it sufficient to include stock variables (more to come) in their analyses and perhaps to measure post-promotion dips in their models to account for the carryover effects of marketing activities. This is one area where academic research has been around for over forty years which could greatly improve industry practice. More sophisticated models are needed to address the dynamics of today’s marketplace, so be sure to challenge your modeling vendor in this regard.
How are advertising effects being measured?
Advertising has long been the focus of most marketing mix analyses. So much so that these are often referred to as media mix models. However, most of these models capture the short-term effects of advertising at best. AdStock (Broadbent, 1979) has become the industry standard for inclusion in marketing mix models. While AdStock is a nice concept to represent the “stock” of current and past advertising levels (i.e, spending or impressions), isn’t it the effect of advertising that carries over as opposed to the effort itself?
Even if one subscribes to the use of AdStock in these models, estimation of advertising effects is not a simple process in doing so. The challenge arises because two parameters need to be estimated at the same time. AdStock itself depends on the decay rate associated with the medium and/or message, while sales response depends on the relationship between sales and AdStock as the measure of advertising. In practice, most analysts treat this in a trial and error fashion or limit themselves to historical rules of thumb for decay rates. When millions of dollars are often on the line, this is not the best approach for measuring advertising effects – even in the short term.
In addition, most marketing mix models ignore the longer-term effects of advertising. Plenty of readers have perhaps seen presentations which report the short-term effects, as modeled above, and then are told that prior research (e.g., AdWorks I) has shown that long-term effects are at least twice those of short-term effects. Since so many of us believe that advertising does indeed have longer-term effects, shouldn’t we be explicitly modeling these effects? This is another area where industry practice has not leveraged academic research nearly enough.
How are the model’s outputs being used?
Pie charts with volume decompositions and “due-to” analyses are standard to nearly every presentation of marketing mix analysis. While these have great visual appeal and some descriptive validity, marketers should exercise caution in how they base decisions off these analyses. Volume decompositions are derived by simulating mode results in the absence of a marketing driver, say TV. The volume associated with TV advertising is then represented as the difference in volume between the two scenarios. The ROI from television advertising is then calculated accordingly and compared against ROIs from other marketing drivers. However, do you really want to look at going from “zero to sixty” like this or shouldn’t you be concentrating on the margin from “sixty to sixty-one”?
In closing, mix models have become prevalent in every industry where there’s a response variable and associated marketing activity to be measured. Most of these models are based on marketing science which has been around since the 1960s. Anyone can look around and see that the marketplace of today is very different from that of the 1960s. Many advances have been made in the academic arena with respect to marketing science, but few in industry seem to be catching on. This is one area where practitioners should learn a lesson or two from recent academic advancements!
Please feel free to contact me at email@example.com with questions or comments.