Chapter 6: Demand-Responsive Supply Chains


Published on

  • Be the first to comment

  • Be the first to like this

No Downloads
Total Views
On Slideshare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Chapter 6: Demand-Responsive Supply Chains

  1. 1. Chapter 6: Demand-Responsive Supply Chains The need to make or order products before demand is known with certainty is one of the basic challenges of supply chain operations. Too much product means that it has to be stored for a long time, incurring inventory carrying costs and likely being sold at a discount or a loss; too little inventory means lost sales and lost customers. Typically, companies use forecasting techniques based on statistical models to predict demand and use the resulting forecast to produce or order the amount they anticipate their customers will demand. But even with the most sophisticated forecasting models, forecasts are inaccurate. And forecasting methods are even less useful for predicting low-probability events. Thus, it is instructive to study how companies facing significantly uncertain demand organize their operations. These companies minimize forecast errors by creating a more responsive supply chain so that they can react to demand fluctuations even when not anticipated. Demand Disruptions In mid-2000, IBM revamped its laptop computer product line, the venerable ThinkPad, releasing the T20 and A20 models. After losing $800 million in 1998 and $571 million in 1999, the IBM PC division used a conservative forecast of the sales of the new machines and kicked off a major ad campaign in conjunction with the release. However, to the delight of the IBM marketing department and to the horror of its operations department, the new ThinkPads became an instant hit with consumers. Sales soared, leading immediately to product shortages. In mid-July, customers seeking 79 of the 108 ThinkPad configurations faced back orders of well over a month. The problem had no quick fix because component suppliers were geared for the original forecast and could not quickly ramp up their production of DVD and CD-RW components. One of the most unfortunate problems with such “under-forecasting” is that IBM does not even know how many potential sales it lost to competitors, as would-be buyers were turned away by the publicity regarding the shortages. Erring on the side of over- estimating demand, however, can also be detrimental, as items may have to be sold at discount, robbing the manufacturer or the retailer of its profit margin and even forcing it to sell at a loss. Such discounts are commonplace in both fashion apparel, which is subject to teenagers’ whims, and consumer electronics products, which lose their appeal when a new model or gadget comes on the market. But even in mature products, like automobiles, discounting is common. For example, in 2004 American manufacturers were offering $3,000–$4,500 rebates on sales of sport-utility vehicles when demand ran below forecast, in part because of high gas prices. In some cases, products flop completely, as was the case with the Ford Edsel. After unprecedented consumer research and an extensive ad campaign, Ford introduced the car on September 4, 1957. It quickly became clear, however, that consumers preferred the silhouette photos in the marketing brochures to the actual cars. Instead of the 200,000 Edsels that Ford was geared to sell, only 63,110 were sold—with discounts. Even after a restyling for the 1959 model
  2. 2. year, only 45,000 Edsels were sold. Ford pulled the plug on the car one month after the 1960 model was introduced, again to disappointing sales. New product failures are not confined to automobiles. In the consumer packaged-goods industry, almost 80 percent of all new products and new product variations fail within the first two years. In the industrial products sector, 30 percent of all new products fail.[1] And the results of such failures can be dramatic. In 2003 two pharmaceutical companies—Wyeth Inc. and MedImmune Inc.—jointly introduced a new flu vaccine that could be inhaled rather than injected. Unfortunately, it was not a consumer favorite even during the flu vaccine shortage of 2003.[2] Wyeth was able to sell fewer than 400,000 FluMist doses out of 4 million produced and had to take a $20 million charge in January 2004 to destroy[3 ]and write off the unsold stock. Wyeth has announced a plant closing and the elimination of several hundred jobs[4] while Med- Immune stayed in the market but produced only a few vaccines in 2004, contributing to the influenza vaccine shortage of 2004.[5] The world of commerce is strewn with products that failed despite sophisticated customer research and lavish ad campaigns. The failure of New Coke, Crystal Pepsi, the Betamax VCR, the Apple Newton, and Microsoft’s Bob demonstrate vividly the difficulties of forecasting. Forecasts’ Characteristics Regardless of the sophistication of the underlying approach, the characteristics shared by all supply chain forecasts include the following: • Inaccuracy The most glaring attribute of all (point) forecasts[6 ]is that they are invariably wrong. This is simply a statistical reality. For example, forecasting the monthly sales of a certain yellow women’s blouse in size 8 at a given price is bound to be wrong because there is a certain probability that it will equal almost any number. Since the forecast is a single number, the probability of the actual sales matching exactly the forecasted demand is practically nil. • Improvement with aggregation A second characteristic is that aggregate forecasts are more accurate than disaggregate fore- casts. Forecasts can be aggregated, for example, over time, geography, or products. With aggregate forecasts, errors tend to cancel each other out, leading to more accurate forecasts.[7] While it is difficult to forecast the sales of a blue men’s blazer size 42R on a given day in a given Boston store, it is easier to forecast the monthly sales of that blazer in that store, and even easier to forecast the monthly sales of that blazer throughout New England. • Time Horizon As anybody who follows weather forecasting knows, long-range forecasts are less accurate than short-range ones since fewer factors are known the longer the time frame is. Likewise, sales trajectories can diverge further and further from a projected forecast as time progresses. New fashions, economic changes, and competitors’ actions make the distant future murkier than the near-term future. But many supply chain operations require long-term demand forecasts since orders involve long lead times. • Reliance on history Forecasting methods use historical data and experience. The competitive environment drives manufacturers and suppliers to introduce new products
  3. 3. and new versions of old products continuously. In these cases, and when companies enter new markets, data are scarce, making it difficult to forecast. • Reliance on trading partners History, however, is not the only source of data; trading partners often have information that can help in forecasting and planning. For example, retailers can give their suppliers data on sales patterns throughout their stores so the suppliers can base their forecasts on actual consumer behavior rather than the retailers’ order pattern. • Risk sharing While the sharing of data may lead to more accurate forecasting, companies can also share the risk of forecasting. Even though this will not improve the forecast itself, the practice can help supply chain partners mitigate the consequences of wrong forecasts and increase the profits of all trading partners. Companies facing uncertain demand are doing more than investing in better forecasting tools. Acknowledging the inherent variability of demand and the limitations of statistical forecasting, they use these characteristics to design their supply chains to be flexible and to respond to ever- changing demand patterns, thus making them less dependent on demand forecasting. The flexibility to respond to demand fluctuations, created by these supply chain designs, also increases these companies’ resilience to disruption—be it an unexpected demand surge or unexpected problem with their supply lines. The next sections describe how companies build flexibility into their operations based on each of these six forecast characteristics. Building Responsiveness with Range Forecasting Instead of forecasting a single demand figure, progressive companies have turned to forecasting a range of potential outcomes. The range is used as a guide for supply contracting terms and contingency plans: what to do if demand is on the high end or low end of the range. More important, the use of range forecasting conditions the company to think in terms of uncertain outcomes or a range of possible realizations. Companies can and do use range forecasting for more than just estimating future demand volume. For example, Agilent, Inc., develops a range forecast not only for future demand but also for supply volumes and prices for all of its products. Agilent is a $6 billion manufacturer of scientific instruments and analysis equipment, spun off from Hewlett-Packard in 1999. After developing all the possible outcomes for each new product, it assigns probabilities to each outcome that has more than a 10 percent likelihood of materializing and develops contingencies for these outcomes. Thus, Agilent’s planning covers 80 percent of the potential outcomes. Range forecasts are used in flexible contracting, procurement strategies, and financial planning. The objective is to increase the company’s flexibility, as the range forecasts prepare the company for changing market conditions. [1] P. Kotler, Marketing Management: Millennium Edition, 10th ed. (Englewood Cliffs, N.J.: Prentice Hall, July 19, 1999).
  4. 4. [2] Linda Loyd, “Wyeth to Close Lancaster County Plant,” Philadelphia Inquirer, March 26, 2004. Retrieved October 10, 2004, from [3 ] Flu vaccines have to be formulated anew every year because the flu strains change from year to year. [4] Michael Barbaro, “FluMist Offered Free to Public Health Agencies.” Washington Post, January 21, 2004. Retrieved November 10, 2004, from dyn/A33882–2004Jan21 [5] Oral testimony of Jim Young, president of research and development at Med- Immune, at the U.S. congressional hearing on government reform on October 8, 2004. Retrieved January 10, 2005, from %20Young%20Vaccine%20Testimony.pdf [6 ] Point forecasts are forecasts constituting a single number, such as the future sales of a given product in a region during a given month. [7] This statement assumes that the errors are independent of each other. When the errors are positively correlated the effect is diminished; when they are negatively correlated the effect is strong.