Eliminate statistical outliers from future forecasting models!
Many demands are mistakes; requests for the wrong item or the wrong quantity. Outlier identification and pattern recognition can and should prevent those demands from entering the system. An outlier is a data point that lies so far outside the norm that it should be tested, called into question for reasonability, and handled differently from normal demand data points. An example outlier is the demand during July, 1995 in the figure below.
I came across an outlier and its effects when I was working with a roofing tile manufacturer in Florida a few years ago. As I toured their warehouses I noticed that they were completely full. As I toured their manufacturing operations I noticed the production lines were running at full capacity with no end in site. I asked the plant manager what their plan was for storing all the tiles. He had no answer. He explained that they were just following the production plan. Then it dawned on me that we were exactly one year down the road from hurricane Andrew, one of the most devastating on record. Can you imagine the impact of a hurricane on the demand for roofing tiles? Someone had not identified the hurricane demand as an outlier and the roofing tile manufacturer was running a production plan based on the impact of one of the most devastating hurricanes on record. That’s an outlier that should have been eliminated!