Forecasting Journey at on Semiconductor
Williams, Tim, The Journal of Business Forecasting
If Sales are entrusted with the responsibility of forecasting, they are more likely to over-forecast, because the higher the forecast number, the larger the capacity reservation they will get ... forecasts don't improve if their performance is not measured ... when serving customers through the distribution channel there is a tendency to over-forecast demand during an upturn if product shipped into the channel is recognized as revenue.
The forecasting function at On Semiconductor was put in place in the 1990s when it was known as the Semiconductor Components, a Division of Motorola's Semiconductor Product Sector. My first encounter with the forecasting function was in 1997 when I was a business manager in a wafer fabrication facility in Phoenix, Arizona. My factory manager just returned from the Operations Review meeting with a bad news: senior management had decided to consolidate fabrication operations and relocate the manufacturing of our products to a sister location in Seremban, Malaysia. He was somewhat disturbed by this action because the demand forecast or Request for Product (RFP) was still extremely high, despite the fact the market for these products was somewhat flattened. He asked me to calculate the required factory capacity in order to support the RFP. I did the analysis, which validated his belief. If our forecast was right, we could not only not afford to shut down our fabrication plant and consolidate operations, but also we had to build an additional facility to meet the demand. Management was correct that the forecast was extremely high compared to actual demand, and in 1998 it shut down the fabrication plant in Arizona. The objective of this article is to share our forecasting journey-where we were, where we are now, and how we got here. We have divided our journey into four phases.
Our forecasting journey begins in the late 1990s with the Request for Product (RFP). At that time, Motorola reorganized itself and became market focused. The Sales Department was given the responsibility to forecast demand for their customer accounts. Knowing how resources are allocated at the time, Sales quickly learned that the higher the forecast number the larger capacity reservation they would get.
At that time, forecast metrics were not used and, as such, there was no incentive to improve forecasts. However, underestimating demand in the preceding years had resulted in lost revenue and dissatisfied customers. This led to extremely biased, high forecasts. The unrealistically high forecasts caused many functional areas such as Manufacturing, Planning, and Finance to discard the RFP numbers for their own functional plans. The Sales Department was not happy with that. It felt that because preparing forecasts was a tedious and time consuming process and yet no one trusted its numbers, it decided to abandon this responsibility altogether.
Phase II starts in August 1999 when Texas Pacific Group completed the largest U.S. technology leveraged buyout of $1.6 billion of the Semiconductor Components Group of Motorola. Within less than a year, our company went public as On Semiconductor. At that point, we had just implemented i2 Technologies Demand Planning Software and were rapidly learning its capabilities. Business Units in the newly formed company quickly came to the forefront and each had strategy groups that ran their business. These groups had little time for collaborating with the forecasting function and often overrode the entire forecasts with their own static business plans that had been developed on spreadsheets. These plans represented stretch business goals, which was quite common at the time because of the growing semiconductor industry.
The forecasting function reported to the Supply Chain organization, which resided in manufacturing operations. It held a monthly consensus meeting in which the statistical forecasts and business plans were reviewed for the major manufacturing resources in each product family. …