Marketing mix modeling (MMM) is an analytic approach, through historical information, that quantifies the impact of marketing elements on sales or market share. This practice has been inexistence since 1949 (Source: Wikipedia) Thanks to the development of technology and the availability of in depth data, MMM has been witnessing a quantum progression in terms of the approach that's been adopted. MMM as a concept was used by consumer packaged goods (CPC) industries only, a few years ago, but in recent times, its popularity is spread to other industries as well. As the penetration of MMM in various categories and markets expanded, the discipline also witnessed advancements. Increasingly MMM is considered a combination of art and science. Domain expertise in marketing and media is one of the necessary factors in successful implementation and new methodologies have exploited this connection
MMM, helps us to quantify the effect of marketing inputs on business metrics- sales/market share. This is usually done through a statistical technique called "regression". Till recently, regression was used by treating most of the marketing variables as linear and media, especially TV as non linear. By treating the influence of TV as non linear, it embraced the theory of diminishing returns on TV deployment- as the input on TV increases the sales/market share increase at an increasing rate in the beginning and then increase at a diminishing rate. This approach helped the user to set an optimal TV weight for a desired sales/market share. Increasingly this approach was followed when TV was considered as a lead medium and also this technique was used for setting TV weights.
As the usage of MMM expanded, the utility of the same also expanded beyond just TV weight setting. The approach found increasing usage in determining the type of distribution outlet that influence sales/market share, determination of optimal price for the brand and the interrelationship of media especially between the paid, owned and earned media on sales [market share movement. Increasingly MMM is being used not only for deriving the insights but as a "live" scenario planner answering "what if" questions to the marketing practitioner.
Auto regression was used as a technique in implementing MMM by many practioners in the past and many continue to do so. This related the relationship of set of independent variables, the marketing inputs, on sales. But the underlying assumption Of this technique was the variables taken in to consideration should be "independent". But in reality, especially in marketing, one does not come across "independent" variables. All variables are interlinked-TV not only could influence sales but also could influence distribution. Similarly online media and TV could be related in a close manner as TV exposure could lead to an online visit and this could then result in brand purchase. So one of the key developments in this discipline in the recent past is treatment of these variables not as "independent" but more realistically as interdependent variables. Techniques like Bayesian network analysis, Bayesian shrinkage, causal model using Bayesian, Vector auto regression (VAR) are some of the advances from autoregression. These techniques provide us insights on the extent of relationship between the so called "independent" variables and thus by embracing the inter relationship, provide us a more precise estimation of the influence of marketing variables on sales. It not only provides us just the estimation, but also throws light on the pattern of direct and indirect influence of certain variables on sales/ market share.
Another major advancement that one witnesses in this discipline is adoption of non linear response functions. Response function is the pattern of relationship of any input variable on sales/ market share. This throws light on the pattern of sales movement as one increase the input on the marketing variable that influences sales. Earlier given the complexity all the other variables apart from TV were considered linear. Now non linearity can be applied to any variable and the complexity has not become manageable, therefore leading to better estimation.
But in reality most of the variables including the digital media, need not display a linear relationship. The relationship could be nonlinear as shown in exhibit two and three
But in reality most of the variables including the digital media, need not display a linear relationship. The relationship could be nonlinear as shown in exhibit two and three
Thus understanding the pattern of relationship and then mirror this pattern in modeling is very critical for an accurate MMM. New techniques like Bayesian network, causality models using Bayesian and Bayesian shrinkage are increasingly used to take care of this pattern.
MMM has evolved from being used as to derive media insight to a robust aid in creating a complete marketing strategy. The output from MMM is increasingly being used to create various scenarios in deploying marketing inputs and monitor the business outcome. Since the capture of data has being automated and is not accessible without a lag, the scenario planner from MMM is used more like "live" scenario! In addition, given the advancements in optimization algorithms, MMM is also used increasingly to optimize the marketing and media inputs to deliver sales growth and marketing effectiveness.
Some other additional insight possible are,
More and more industries are embracing this practice in recent times, as marketing accountability is one of the core responsibilities of a marketing practitioner. MMM has moved a long way from being used by CPG category to even to most of service categories. Irrespective of categories, getting a causal insight, through quantitative techniques from the available data is becoming a basic norm for marketers. As this discipline penetrates to many categories, more ideas will get injected in to MMM practice, thanks to category and market dynamics. Thus one could see further advancement in this practice in future!