Next-generation Revenue Forecasting and the Importance of Collaboration

Monday, March 19, 2018

A revenue forecasting system we delivered for a publicly traded North American multimodal transportation company, summarized in a case study on this page, is noteworthy for many reasons. In this post, we focus on the collaboration throughout the project that helped the business team build confidence in changing forecasting processes from a manually intensive methodology to an automated system that provides insights into the underlying forecasting techniques. While many software vendors are promoting automated machine learning algorithms, we believe that the best predictive analytics blends business experience and insight with advanced analytics techniques.

Previously at the transportation company, different teams manually created quarterly forecasts. Each team was responsible for a category of transported commodities such as agriculture, metals, paper, chemicals and industrial products.

A Princeton Consultants team helped standardize the methodology for these quarterly forecasts, using a variety of forecasting techniques, and reduced the time to produce the forecasts by thousands of man-hours per year. Following are key steps in the project:

  1. Internal subject matter experts and external economists were interviewed to build a broad portfolio of indicators specific to each forecasted commodity. As a result, the stakeholders gained confidence in the new system because they provided the input to the key drivers—as opposed to simply relying on a black box machine learning algorithm.
  2. The system produced confidence intervals rather than point forecasts for each commodity, and business executives were trained to understand how these confidence intervals correlated to uncertainty in the forecasts.
  3. Different models were built and compared, along with a scoring system for the business to rank the models within each category. The users learned how the scores correlated to different aspects of model quality and could use the ensemble of models to gain confidence in the forecasts.
  4. The models were designed to accommodate manual adjustments based on non-economic events such as the addition of new customers. This allowed the business to provide qualitative inputs that would influence the forecasts.
  5. The system created multiple scenarios of forecasted freight—such as Stronger Near-Term Growth, Next-Cycle Recession, Stagflation, Below-Trend Long-Term Growth, and Low Oil Price—leveraging alternative scenarios to the baseline forecast each month and often creating scenarios around predicted events. These comparative scenarios increased trust in the forecasting system.

Through a collaborative approach, the Princeton Consultants team helped develop a system that produces superior, faster forecasts that the business users accept and trust. Read more in this case study.

Contact us about revenue forecasting and to explore how we can help you apply predictive analytics to your business problems.