Market Mix Modeling for Dummies

Market Mix Modeling is the estimation of the incremental increase in sales caused by a marketing expenditures such as couponing, advertising, promotional price reductions, direct mailings, telephone solicitation, etc.

The purpose of the process is to understand the relative effectiveness of expenditure A vs. B vs. C so that the marketer can do more of what works best and less of what works relatively poorly.

The “modeling” is done by looking backward at business results over a period of time and attributing differences in sales to the different expenditures. Therefore, in order to make a proper estimate, the modeler must be able to separate the stimuli either in time or sales area.

For example, the modeler must have data with a coupon in one period and not another, or media in one place and not another. Overlapping of various efforts can be sorted out just as long as some differences in time and geography exist to provide base periods and control areas. Identifying these base and control periods and areas is one hurdle faced by modelers.

Once the base and control periods are identified, the modeler needs historical data over a period of time. Modelers call this data “time series” data indicating its availability over sufficient periods to smooth out things such as seasonality. The modeler then attempts to explain the variation in the results in the test area vs. the control area. A mathematical technique known as “regression analysis” is used. The purpose of this technique is to find a number (a “co-efficient”) that can explain the difference in sales between the test area and the control area (or period). In effect, the modeler is doing repetitive “trial and error” calculations trying to find the number that “fits” or explains the differences between the test and control areas.

In theory, the larger the “co-efficient”, the larger the effect on sales of the marketing investment. The co-efficient (a number like 2.1 or 1.8) is adjusted for variation in the cost of the vehicle, so that one is comparing the effect of $1-million of couponing vs. a similar $1-million expenditure in price reductions. The final output of a market mix model analysis is a co-efficient for each of the various marketing expenditures. Comparing these simple numbers allows one to understand which marketing expenditure works most efficiently.

FAQ’s about MMM

1. Where does the data come from?
   Usable “time series” data can come from a variety of sources: the manufacturer’s shipment data, retail store POS data, household panel data, retail loyalty card data, etc. The real issue is having data in sufficient quantity from the appropriate time periods and geographies.
2. Can you measure any marketing expenditure?

   Yes, if you have enough data points from enough locations or individual consumers. If, however, a brand spent a small amount of money on a 500-person test mailing, marketing mix modeling is probably inappropriate because the limited sample size would not permit a statistically reliable analysis.

3. What are the biggest criticisms of Market Mix Modeling?

   It’s backward looking, so it captures a perfect picture of a world that may not exist today or ever again. Changing your ad copy could make next month’s ad expenditure much more or less efficient than last month’s behind different copy. A competitive brand price reduction initiative could change the price relationship implicit in the backward-looking MMM. Additionally, difficulty in separating data into clean test and control areas/periods can cause modelers problems in interpretation.

4. Can’t I use MMM as a guide to future expenditures?

   Sure. That’s why marketers invest in them and use them.

5. It sounds like MMM is "predicting the past"?

   No. Market mix modeling is explaining the relative effect of past actions in causing marketplace variation.