How to Reduce Your Forecasting Error by 35%
This article presents a clear step-by-step approach to reducing your forecasting by up to 35%: collect and clean your data, use machine learning, and enrich your forecasts using the forecast value added framework.
Poor forecasts result in excess inventory, obsolete products, inappropriate long-term capacity planning, supplier tensions, shortages, and emergency expeditions. In short, they harm profitability, frustrate your teams, and waste resources. On the other hand, accurate forecasts will allow you to produce more effectively the required products, ship them on time where they are needed, improve long-term supply planning, and increase your service levels while reducing your inventory levels.
Thanks to our experience creating models, training, and coaching professionals, we could boil down our technique to improve forecasting accuracy into three main steps, as highlighted in the figure below.
Step 1: Data Collection
This first step focuses on data collection and cleaning. It is often overlooked as it is not the sexiest, but it is the required foundation for your forecast-improvement project. Without proper execution of this crucial step, there is no hope that you can succeed with your overall objective.
The most critical piece of data to collect is historical unconstrained demand. Unfortunately, most supply chains are tracking constrained sales instead. (See this article for more information about the difference between the two.) The easiest way to unconstraint your sales is to track historical inventory levels and censor periods with shortages in your forecasting engine. (Some statistical models can do that, but, unfortunately, most software vendors do not include this feature.)
Note that to grasp the impact of shortages fully, you should forecast weekly or daily demand.