Marketing Mix Modeling: Driver Analysis

One of the key modeling concepts used extensively by brands globally is the concept of Marketing Mix Modeling (MMM). First looked at by brands as far back as the 1980’s – it became a force in the late 1990’s and today is de rigueur for most of the big brands. Basically, MMM uses a statistical concept called “regression” to understand the impact of inputs (advertising, pricing, promotions) on the expected outputs (sales, brand perception scores) and quantify the relationship. Basically regression is nothing but changes in output for a unit change in input, collated over many time intervals (months, weeks), across many variables. The basic output is a driver chart as shown (above).

Let us understand what this means.  The sales for the period modeled according to the model are explained as follows. TV accounts for 30% of the sales for this period which if I assume was 1000 then the contribution of TV advertising to sales is 300. That is easy to understand and the same applies to all the positive contributors above. 35% of the sales are attributed to something called the base sales – which is to be read as sales due to accumulated equity and other unexplained positive factors not applied in the modeling period. The chart above also shows some negatives for example competitive advertising. This means that in our example sales of 100 were lost due to competitors advertising and if there was no such advertising; the sales would have been 1100.

We get asked a lot how it is possible to estimate separately the impact of these activities when in fact many of them are running concurrently. While it is impossible for the human mind to actually separate this given the number of patterns that need to be studied, computing power aligned to statistical software can do this. This is the reason why hitherto unknowable insights can now be sought!!

Once we have the basic drivers quantified, a lot of issues that have dogged marketing disappear. Now the marketer knows with high level of statistical significance (in other words, less chance of being wrong) what is driving the sales increase. The working out of the ROI of each media/input is just one step away, as the sales % due to TV or print can be converted to volume sales and return on investment is value sales generated divided by the cost of media deployed. It’s that simple.

What therefore are the some advantages of MMM over other traditional methods of campaign evaluation?

  • MMM is based on actual behavior that is an increase or decrease in sales due to specific stimulus as opposed to a typical media evaluation that is based on likely viewers and then the impact generated by these viewers on sales. In any case Media evaluation is only for media variables whereas  other factors like price, distributions and promotions need to be studied to get an overall picture
  • Compare different variable spends with the same yard stick – return on rupees spent. This is not otherwise possible.
  • Weed out underperforming channels and redeploy the money to better performing channels
  • Deep dive into each of the variables for further insight ( and that’s another story)

MMM has become a stock in trade for many practitioners of marketing, in fact not using it could be akin to a detective not using DNA and finger print data. The methodology is flexible and can be used across manymarketing issues to gain insight.

The disadvantage – has to be used by trained practitioners with domain knowledge as otherwise it can just wrong and sometimes dangerously wrong. More in the next salvo!

Published with permission from RainMan Consulting


Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.