All models are wrong, but some are useful.

George E. P. Box

                                                 

Let’s face it: All models are wrong, but some are useful. The word “modeling” has become very popular since the beginning of the coronavirus in the US. Dr. Deborah Birx, one of the key members at the White House task force, in one media debrief meeting, spent around 10 minutes talking about the models used to predict coronavirus spread and death, and how different actions could change the outlook. If interested, see the video clip here. 

White House task force models potential death toll of coronavirus 

Dr. Birx has a team of more than 20 data scientists to run the models. Now it’s two months later. If we look back to the models, did she or her team predict accurately? Not really. 

And that’s what makes it a model. A model is nothing but a good approximation of what might be happening in the future, IF the assumptions satisfy. Modeling is just a simplified version of the problem being studied. Therefore, all models are wrong. 

Yet, we still need models. We need models because if a simplified version won’t help us understand what’s going on with your problem, it would be harder to understand the real messy situation. Models are useful. 

Marketing Mix Modeling (MMM) 

Marketing Mix Modeling (MMM) is a technique that uses quantitative methods to measure the impact of marketing inputs on sales or market share. The main purpose of using a marketing mix model is to quantify how much each marketing spending contributes to the sales, and to determine how much to spend on each input. MMM helps marketers in not only determine the effectiveness of each input, but also in optimizing marketing mix therefore maximizing ROI.  It is a tool that is commonly used in marketing budget allocation. 

MMM usually uses regression analysis and the model built through Regression is used for extracting key insights. There are many regression techniques to fit into the problem we are trying to solve. Depending on the type of purposes, linear regression and logistic regression are probably the most commonly used techniques in marketing mix models.

Here is what a simple regression equation looks like. 

Y=a+bX

X= (x1, x2, x3…), it means independent variables.  They can be print ad spending, google ad spending, trade show spending, and etc.  Y is called a dependent variable, which is usually sales or market share. 

You will need to collect data for a certain period of time. For example, if you decide to collect data on a weekly basis, you may need at least one year. That will give you 52 data points to build a basic model. A free software such as R can run regression analysis. There are lots of nuances in model selections, assumptions and explanations, so leave to a pro if you don’t feel comfortable. 

That’s my quick and short brief about marketing mix modeling. In today’s marketing, we need data driven decisions more than ever. Not all beautiful models will give you a solution in the real world, but it’s better to start with something that makes sense. Something better than a guess is better than a guess. 

Give it a try.