Sales Commission Forecast for a Social Media Company

“In 1943, Thomas Watson, Chairman of IBM, said that there is a world market may be for five computers” (“LinkedIn Pulse (2016)  P1”). Steve Balmer predicted that iPhone would never get a significant market share (“Business.insider.com (2012) P1”). Thomas Watson was somewhat right, but Steven Balmer was utterly wrong. 

Forecasting has traditionally been gut-feeling. Truly when someone decides based on gut feeling, they quickly crunch the past perception about the problem (derived based on experience on past data) and promptly decide in an unstructured manner. Modern-day forecasting brings science to a gut-feeling job in a structured way. 

Forecasting is a barometer for the company, and it quantifies the direction that the company should steer. Forecasting helps understand the spikes and turning points and can uncover opportunities for making the right decisions. There are multiple Forecasting use cases in the corporate world, from revenue forecasting, headcount forecasting, booking forecasting, customer payment forecasting, customer write-off forecasting, etc.  

Forecasting the Sales Commission is one of the most challenging tasks for any sales finance organization due to the variation in unpredictable sales revenue, headcount growth, etc., which are the drivers for sales commission forecasting.  

The objective of this case study is to forecast the Sales Commission for a social media company with the best-performing model with the highest accuracy and lowest error. Forecasting is a time series problem, and past data for 48 months helps to train the model. 

Various techniques, including ARIMA and SARIMA, address the time-series problem. Also, supervised machine learning techniques can solve the time-series problem using lag features. LSTM is another advanced deep-learning technique useful to solve sequenced time-series data. This project will use all these techniques to forecast to arrive at the best forecasting model. 

The forecast accuracy results help pick the best-performing model for ongoing forecasting. Also, various fine-tuning methods to improve the model performance and accuracy are helpful for the model monitoring process.