Statistical Analysis

Trade promotion evaluation systems focus on the process, potential benefits, types of information and reports.  A successful trade promotion system/analysis must address these items but the most critical item is an accurate and believable quantification of incremental profits and sales.  If incremental sales and profits are not accurate or not believed then everything else can be 100% perfect and the system will fail.

No model is perfect, no set of data 100% accurate and unknown activities impact sales.  An analyst needs to be confident the results are accurate and questions from management can be answered (not answering reasonable questions creates doubt about the results).  An analyst needs a statistical overview of what the model has found.

This analysis is more important for the more sophisticated models.  Simplistic models such as smoothing models or even univariate models will not be reliable.  The principles applied below work for multivariate models and while they can be applied to univariate models but it is important to remember the shortfalls of univariate models in calculating various elasticity's.

 Here are some principles for that overview:


  1. The estimate for the SPARLINE (baseline) needs to broken into how much each marketing condition contributes.  One of the most common questions is why does the SPARLINE equals x.  By having monthly estimates the analyst can show which factors were considered and how much each one contributed to the SPARLINE.  This allows for an objective discussion as opposed to a broad challenge about the accuracy of the numbers.
  2.  Quantify the elasticities of the marketing variables.  When analyzing thousands of promotions,  illogical conclusions may result.  This usually occurs because something unknown happened in the market (e.g. a snowstorm reduces sales).  This results in elasticity's that have the wrong sign or an illogical magnitude.  For example a coupon is dropped and the elasticity (or correlation) between the coupon and sales is negative.  Remember that correlation is not causation.  During the coupon period there may have been unusual weather, other factors such as OOS or competitive activity to drive down sales.  If these factors are not built into the model, it could result in an illogical elasticity since it is not logical to show that coupons reduce sales.  The analyst needs to see these elasticities to make sure they are logical.  If they are not logical then more research can be done to make the analysis better.
  3. Forecast the future.  After successfully analyzing the past the next question is can you predict the future?  An analyst needs to see the projections in the same format as the history to make sure they make sense.