Analytic Forensics Logo 






I have spent over 20 years creating Marketing Mix Models for Fortune 500 companies and the government, and I also taught Marketing Mix Modeling classes.  Since not everyone knows what Marketing Mix Modeling is, I’ll explain it below (this material was taken from my Marketing Mix Model class):

Marketing Mix Modeling is the science of isolating the effects of advertising and marketing components on sales.  Often these effects are presented in terms of Return on Investment (ROI).  This is done using “econometrics” (a statistical technique that has been popular for the last 120 years) which uses history to translates complex dynamics (like the economy but also things like marketing and sales) into equations.

The goal of marketing mix models is a chart like the one below (among other things).  By finding the ROI of each marketing factor you can tell which marketing or media are the most effective and the least effective.  For example, an ROI of “3” means you get $3 in revenue for every $1 you spend.   You can also use it to predict future sales, and optimize the best level of marketing spending.

How this is done (to over-simplify) is to look at the historical “peaks and valleys” of ALL your marketing programs over time and compare them to the peaks and valleys of your sales.  Look at the graph below – It shows an example of the monthly ad spending (TV, Print, and Promos) compared to sales of a brand.

The basis of Marketing Mix Modeling is simply looking at all the marketing spending and all the sales and just seeing how the two relate.  You don’t ask why the sales peak, often the “why” is a complex mix of many factors.  You just look for correlation and you turn that into ROI.

When looking at each marketing the process is basically the same principle.  You look at individual marketing factors (like TV spending in the graph above) and look how it relates to peaks in the sales.  That relationship shows contribution of that factor.  For example, in this graph below, you can see that sales peaked 2-3 months after each big TV buy.


Of course, the effects (sales peaks) of each marketing component tends to hide each other, so to find all the correlations, you need fancy statistical techniques.  Usually you use least square regression, but using Bayesian Shrinkage models are also common, and there are many other techniques.  How you create the equations is not important, what makes it a Marketing Mix Model is finding the correlations from spending on marketing factors to sales and expressing that as an equation.

Now this was an over-simplification.  The relationship between marketing and sales is a complex system of effects.  In Marketing Mix Modeling we try to look for the simple direct correlations at the core of all that complexity, but to do that, you need to put some of that complexity into your models.

The biggest complexity you need to deal with is “time”.  Marketing Mix models are said to be “time series” models.  This is because an ad never leads directly to a sale.  Marketing is meant to build awareness, encourage consideration, and change brand perceptions.  And all those changes in the consumers eventually translates into sales.  But these changes take time, so you see the effects of an ad over time and often “peaking” and then trailing off.  This effect over time is called “ad stock” (click here for more discussion of Ad Stock).  Ad Stock is different for every product and every marketing factor and calculating it is the hardest part of Marketing Mix Models.

But there are many other non-linear complexities that need to be built into Marketing Mix Models (click here for more discussion on some these complex Non-linear Marketing Effects).

The real art of Marketing Mix Models is balancing the simplicity with the complexities.  If you make your models too simple they are neither very accurate or very useful.  If you make them too complex they became as hard to use and confusing as the marketing effects you are trying to explain.  The perfect balance makes the models the most useful to the client.

But when you find that perfect balance, you can use the models to pick the perfect spending level either for each marketing factor or for your total budget.  You can create simulators that lets the clients create and test “what-if” scenarios.  You can predict future sales.  And most important, you can get the maximum sales return (ROI) for every marketing dollar spent.  With very large corporations this could save them hundreds of millions of dollars per year.


Home Page


Analytic Forensics Logo




Text Box:                                                                                                                                                To Contact Us:







                                                                                                                      Phone:                                                     212-529-5337

                                                                                                                      Voice Mail:                                               917-838-7966


                                                                                                                      Address:                              23 East Tenth Street #304

                                                                                                                                                                New York City, NY     10003