Between 2009 and 2011 I worked as a statistical modeling consultant for Netapp, a global provider of enterprise class data storage solutions. During this time I developed statistical simulation models for sales commissions forecasting which continue to directly influence the budgetary allocations of millions of dollars. Commissions expenses cannot be known in advance because the future performance of the salesforce is itself unknowable. This creates uncertainty during the allocation of funds to pay such expenses. Putting too much money aside carries an opportunity cost of not applying the excess funds more impactfully elsewhere. Conversely, allocating too little money means having to reallocate later, taking the difference from some other budget area.

Due to client confidentiality agreements, I cannot disclose the particulars of my models. However, I can comment broadly on the tools and methodologies employed therein.

First, I categorized various sub-populations within the global salesforce by sales territory and by sales role. I mapped individual sales agents to their compensation plans. Then, I implemented propagation logic to calculate commissions correctly for secondary and tertiary sales agents (such as managers and support engineers), whose commissions are based upon the performance of sales agents other than themselves.

Using distribution-fitting algorithms, I then classified the performance metrics of primary sales agents within each population as being represented by the best fitting distribution type. Distribution types were normal for approximately half of the populations. The other half were best fit by various forms of non-normal distributions, including log-normal, Poisson, and others. Interestingly, the distribution type best fitting certain populations sometimes changed from year to year, perhaps due to shifting market and competitive conditions.

Being able to use a representative distribution type to generate synthetic performance data for a given population of primary sales actors allowed me to proceed with Monte Carlo simulations. These simulations, coupled with the logic to propagate commissions calculations from primary to secondary and tertiary sales agents resulted in a highly accurate forecasting model. Geographically granular salesforce commission expenses for the coming year could now be predicted with accuracy rates of approximately 95% on average. These forecasts, in turn, enabled highly accurate budgeting processes, freeing up millions of previously underutilized dollars for more effective allocation elsewhere in the business.