Enterprise Demand Sensing in the Automotive Industry

I was pleasantly surprised when Professor Chaman Jain asked me to devote this column to my perspective on the evolution of the forecasting function over the past 25 years to commemorate the Journal of Business Forecasting's 25th anniversary. I was mostly surprised, however, because I realized that my experience actually started earlier (30 years ago), in 1976 during my first industry job at a consulting firm, A.D. Little (ADL). This firm was unique because Robert G. (Bob) Brown walked the ADL halls before me and left behind a strong legacy that was passed on to me. He was the person in the late 1950s who was largely responsible for bringing exponential smoothing and other smoothing forecasting methods to industry, especially for use in inventory management. My bottom line on the evolution during my tenure is that forecasting functions got much more important in most companies, and while the environment in which they needed to operate got much more complicated, advanced technology saved the day for everyday forecasters. WHAT DO THE NUMBERS SHOW? To start my research for this column, I pulled out a textbook I used in 1982 to teach a forecasting course. It was the initial publication of Forecasting: Methods and Applications by Spyros Makridakis and Steven C. Wheelwright, published in 1978. In it I found data drawn from surveys done in 1975 and 1976 on the status of forecasting in business firms. I did a comparison of these findings with Professor Jain's 2001 benchmarking findings as published in his book, Benchmarking Forecasting Practices. Here is what I found. The one thing that is clear is that forecasts are revised more frequently than in 1975, as shown in Table 1. The majority of companies in 2001 revised forecasts on a monthly basis in contrast to the majority in 1975 revising them quarterly or longer. The 1975 survey also showed that only 61 and 55 percent of Manufacturing and Inventory Control departments used sales forecasts, respectively. The 2001 benchmarking survey did not ask this type of question, but it did show that 76% of the respondents held cross-functional consensus meetings; e.g., Sales and Operations Planning (S&OP) meetings. This implies that at least 14% more of Operations departments use sales forecasts than they did in the 1975. Lastly, survey comparisons show that companies were using non-mathematical techniques-such as opinion and judgment-based forecasting-more extensively in the past versus the extensive use of mathematical methods today. (see Table 2) INTERPRETATION AND PERSONAL OBSERVATIONS These survey findings are relatively consistent with my observations on the evolution of the forecast function over my career. I've seen that companies are forecasting more often, using the forecasts more to drive operational planning, and using more fact-based quantitative methods in generating forecasts. The driving forces that have shaped these and other changes in the forecasting function over the past few decades include: * Push to Pull Manufacturing: Probably the most significant changes in the forecasting function have been driven by the move from the Push to the Pull Manufacturing paradigm. Up to the early 1980s, large-scale manufacturing companies with strong brands ruled the roost and 'pushed' products out. If they made something it would sell. So manufacturing plants made big production runs of items to keep production costs low, recognizing that everything they made would automatically sell or, if not, the marketing folks would easily get customers to buy what was left over. However, over the past few decades, the increasing affluence of U.S. consumers gave rise to consumerism, which shifted market power towards the retailers and away from the producers. This resulted in manufacturing increasingly being driven by the 'pull' from retailer and consumer demands. For example, P&G is moving more towards the Consumer-Driven Supply Network (CDSN) concept, while AMR Research describes this as an industry trend towards the Demand-Driven Supply Network (DDSN). …