Company

US alternate channels division of multinational Fortune 500 food company

Reduced Forecast Error

Situation

• The company was generating statistical forecasts at material (item) level. 
• Certain market segments and products had significant seasonality.  The most extreme seasonality was in Food Service products sold for use in school cafeterias.
• Production Capacity was highly utilized on an annual basis, requiring significant prebuild for demand peaks during the back-to-school period.
• The company was aggressively expanding school offerings, with several new products and line extensions every school year. 
• Nutritional regulations were changing, requiring significant formulation changes each year resulting in new product numbers in SAP due to the reformulations.
• To generate statistical forecasts on new products and line extensions, the company used like modeling and life cycle planning in APO.  Even with like modeling and life-cycle planning, statistical forecasts for many line extensions were not accurately reflecting seasonality.
• The breakdown between flavors also changed significantly in certain product families with the introduction of line extensions for the new school year.  Line extensions cannibalized existing products, even while some product families were experiencing significant growth.

Solutions

• Developed an Excel tool using pivot charts to visually analyze historical time series data together with statistical tests to identify items that had common historical demand patterns. These forecast groupings did not align to any existing hierarchies in planning software.
• Trained planners to use the Excel tool to identify groupings and to create selection criteria within planning software to generate statistical forecasts in aggregate for these forecast groupings.
• These statistical forecasts were used as inputs to consensus forecasts and Sales and Operations Planning (S&OP). 

Benefits

• Within two months, we were seeing significant improvements in forecast error.
• The impact on forecast error varied for products that we combined into groups varied from slight improvement to significant improvement.  
• The most dramatic increase was a group where MAPE decreased within a year from 38% to 27%.  
• The overall impact to the forecast accuracy in these channels was significant enough that company incorporated the grouping strategy as part of a transition to new statistical modelling software.