Homework answers / question archive / HBR

HBR

Business

Share With

HBR.ORG For the exclusive use of M. Gipson, 2020. November 2012 reprinT R1211J Which Products Should You Stock? A new technique to help retailers improve assortment planning by Marshall Fisher and Ramnath Vaidyanathan This document is authorized for use only by Mika Gipson in 2020. For the exclusive use of M. Gipson, 2020. 2 Harvard Business Review November 2012 This document is authorized for use only by Mika Gipson in 2020. For the exclusive use of M. Gipson, 2020. For article reprints call 800-988-0886 or 617-783-7500, or visit hbr.org Which Products Should You Stock? A new technique to help retailers improve assortment planning by Marshall Fisher and Ramnath Vaidyanathan Illustration: Jenny Bowers G etting product assortment right isn’t easy, yet it’s absolutely critical to retail success. Unlike inventory management and pricing, where retailers have lots of data and analytical tools to guide decision making, assortment optimization is still much more art than science. And making the wrong call can be disastrous. Consider these examples: • Following a survey in which customers said they’d like less cluttered stores, Walmart introduced Project Impact, in 2008, removing 15% of the SKUs it carried. Sales declined significantly, and it was forced to roll back most of the changes. Copyright © 2012 Harvard Business School Publishing Corporation. All rights reserved. November 2012 Harvard Business Review 3 This document is authorized for use only by Mika Gipson in 2020. For the exclusive use of M. Gipson, 2020. Which Products Should You Stock? • Super Fresh, owned by the grocery retailer A&P, stopped carrying many of its low-selling dry grocery items to allow for an expansion of fresh offerings. But the eliminated products turned out to be essential to customers; when they couldn’t find them, they took their business elsewhere, and the retailer entered bankruptcy. • A retailer of home goods used demographic data to localize its assortments to better cater to customers’ tastes. It started with fashion bedding and was thrilled to see an 18% revenue lift. But when it applied the data to the fashion bath category, revenues didn’t budge. Discouraged, the retailer abandoned the effort. • When the new CEO of a tire retailer shifted its assortment from low-priced tires to more-expensive ones, he learned the hard way that price mattered to his customers. The CEO was replaced after two years, and his successor restored most of the products that had been eliminated. Like so many assortment-strategy shifts, these moves were largely acts of faith. It’s easy to spot the dogs in your assortment, of course—sales data will tell you that—but it’s far from obvious what slow sellers should be replaced with. And there is always the nagging concern that a slow seller you delete might be an important product to some of your best customers, prompting them to defect to competitors. As all retailers know, picking the best assortment is a balancing act; any one change can have ripple effects. Plenty of software tools claim to support assortment planning, by helping retailers decide which combination of products will maximize sales. But with very few exceptions, they lack the ability to forecast demand for new products or to estimate how much demand would transfer to other products if a slow seller were dropped. The tools do little more than facilitate a manual planning process that relies on the judgment of managers for key inputs. They do nothing to reduce the risk inherent in every productassortment decision. To address this deficiency, we’ve developed a technique that makes assortment planning vastly more scientific. It is rooted in our observation that most of the time customers don’t buy products; they buy a bundle of attributes. Think about the last time you bought a TV. Did you say, “I want TV X”? Or did you think about screen size, resolution, price, LCD versus plasma, and brand? Our approach uses sales of existing products to estimate the demand for their various attributes and then uses those es4 Harvard Business Review November 2012 timates to forecast the demand for potential new products. Armed with these data, retailers can test their hunches more scientifically. Our approach is especially useful for retailers in the hard-goods and grocery segments; it’s less applicable in the fashion-sensitive apparel segment, where products change fast. Grocery retailers currently use the abundance of available market data to identify potential additions—SKUs they don’t carry that sell well at other retailers. But research we conducted shows that our attribute-based approach has a lower margin of error. It also helps retailers gain insight into the following questions: • Can we improve our assortment by replacing slow-selling products with new ones? What is the likely demand for the potential new items? • If customers don’t find their ideal product, what is the likelihood that they will substitute another? • How will sales change if we increase or decrease the number of products we carry? • Does localization—customizing assortments by store or store cluster—make sense? If so, for which categories? If we decide to create clusters of stores with distinct assortments, how many should we create, and what criterion should we use in creating them? By focusing on the attributes of products, retailers can maximize the number of customers who say either “That’s exactly what I want” or “This product may not be what I’d ideally like, but it’s close enough, and I’ll buy it.” Let’s now look at the process, using two examples in auto parts retailing: the tire department of one chain (a research project we conducted) and the appearance-chemicals division of another (a consulting job). While the process is described here step-by-step, it is in fact multidimensional and highly iterative; much of the analysis is handled by a computer model, which produces the final recommendations. Understand Which Attributes Matter Most to Customers Using our method still requires some judgment about which attributes are important to consumers and how those preferences might influence their purchase decisions if they don’t find their first choice. The steps below can help retailers tackle those questions. Identify which attributes are important to customers. Most retailers already think about their products in terms of attributes and can readily iden- This document is authorized for use only by Mika Gipson in 2020. For the exclusive use of M. Gipson, 2020. For article reprints call 800-988-0886 or 617-783-7500, or visit hbr.org Idea in Brief When figuring out which existing products to drop and which new ones to add in their stores, retailers still largely rely on judgment. The result is often disappointing and sometimes disastrous. A new technique makes assortment planning much more scientific. It uses sales of existing products to estimate the demand for their various attributes—for example, TVs’ screen size, resolution, price, and brand. It then uses those estimates to forecast the demand for existing products and for others that could be added to the assortment. tify those that matter in their category. They might include price, brand, size, flavor, and color. When we began our research project on tires, the retailer’s category manager told us that the important attributes for tires were brand, the mileage warranty, and size. The retailer offered several nationally advertised brands that the manager believed customers regarded as interchangeable. We grouped those together as National Brands. The retailer also offered three house brands of varying quality and price, which we’ll call House 1 (the premium brand), House 2 (mid-level), and House 3 (low-end). A number of mileage warranties were offered, but the retailer believed that consumers considered many to be equivalent. Therefore, we grouped the mileage warranties into three levels: Low (15,000 to 40,000 miles), Medium (40,001 to 60,000 miles), and High (greater than 60,000 miles). The four brands and three mileage-warranty levels implied 12 brand-warranty combinations that the retailer theoretically could offer, but some made little sense, such as a high-mileage warranty on a low-priced brand. Only six combinations were actually offered (in decreasing order of quality): National High, National Medium, House 1 High, House 2 High, House 2 Medium, and House 3 Low. The third key attribute for tires, size, includes type (for instance, radial), and whether it is for a passenger car or some other kind of vehicle. Tires come in 64 sizes, which means that there were 384 possible SKUs the retailer could have carried (64 sizes × 6 brand-warranty combinations). But it carried only 105 in most stores. The count varied across the chain, mostly according to the size of the store. The assortments varied as well, but most stores carried the most popular SKUs. Account for what customers will do if you don’t offer their preferred product. Customers’ willingness to buy another product if they don’t find This technique helps retailers do a better job of replacing slow sellers with new offerings, understanding whether customers are likely to settle for another choice if they don’t find their ideal product, and customizing assortments for individual stores or clusters of stores. It’s easy to spot the dogs in your assortment, but it’s far from obvious what slow sellers should be replaced with. their first choice is a crucial input when a retailer considers dropping items. Their willingness depends greatly on the attribute. Customers probably won’t substitute one dress size for another, but they might buy the blue one if red is not in the assortment. Similarly, people are not going to buy a 14-inch tire for a 15-inch wheel, but they might substitute one brand and mileage-warranty combination for another. So in building an assortment, retailers need to account for the fact that if customers don’t find their ideal item, some of them will buy the next-best option and some won’t. In our tire example, we were interested in the percentage of customers that would shift up by one quality level if their first choice were unavailable and the percentage that would shift down. Analyze Current and Potential Sales By Attribute Now we’ll figure out how well items you don’t currently carry would sell and how adding them to your product mix would affect overall sales. This is where the science comes in. Assemble sales data for a recent period. Start with what you know: unit sales of the SKUs you currently carry and each brand-warranty combination’s share of your total sales. This is the foundation of the model. We typically look at six months’ to a year’s worth of recent data. In the tire project, we assembled sales data by SKU for every store over a recent six-month period. November 2012 Harvard Business Review 5 This document is authorized for use only by Mika Gipson in 2020. For the exclusive use of M. Gipson, 2020. Which Products Should You Stock? Figure 1: Analyze Sales Data brand-mileage warranty combination House 1 HIGH House 2 HIGH totals House 2 House 3 Total MEDIUM LOW units sold A 100 29 28 190 B 282 21 30 203 C 11 12 86 D 53 50 284 E 72 64 20 172 570 F 59 97 285 763 G 10 16 14 76 H 7 33 157 377 I 10 183 524 J 39 225 568 K 8 10 73 L 8 47 223 M 43 298 N 72 221 O 8 200 sizes Customers don’t buy products; they buy attributes. A tire retailer we worked with analyzed its sales data according to what mattered most to its customers: size, brand, and mileage warranty. National National HIGH MEDIUM Total units sold Share of total sales 531 7.7% 141 2.0% 182 2.6% 1,331 19.2% Figure 1, showing one store’s data for 15 of the 64 tire sizes, represents the raw data for our analysis. (The data have been changed to protect proprietary information.) Forecast demand for all potential SKUs. The fact that some SKUs had only single-digit sales suggested that replacing them could increase revenue, 3,986 57.5% 760 11.0% 347 536 109 387 898 1,204 116 574 717 832 91 278 341 293 208 6,931 but the challenge was to figure out which new SKUs would sell better. The first step is to use sales data to forecast total demand for each tire size if all brandwarranty combinations were offered. To illustrate, let’s look at size F. (See Figure 2.) Notice that the retailer currently carries size F in four of the six brand-warranty combinations. We start by Figure 2: Estimate Total Demand brand-mileage warranty combination as an example, look at size f: House 1 HIGH House 2 HIGH totals House 2 House 3 Total Share of demand Total MEDIUM LOW units sold captured demand A 100 29 28 190 B 282 21 30 203 C 11 12 86 D 53 50 284 E 72 64 20 172 570 F 59 97 285 763 G 10 16 14 76 H 7 33 157 377 I 10 183 524 J 39 225 568 K 8 10 73 L 8 47 223 M 43 298 N 72 221 O 8 200 sizes The retailer next determined what total demand would look like if all brand-warranty combinations were offered in each size. National National HIGH MEDIUM 1,204÷ 87%= 1,384 total share total sales captured demand for size F for size F Total units sold 531 141 182 Share of total 7.7% 2.0% 2.6% sales 6 Harvard Business Review November 2012 1,331 19.2% 3,986 57.5% This document is authorized for use only by Mika Gipson in 2020. 760 11.0% 347 536 109 387 898 1,204 116 574 717 832 91 278 341 293 208 95.3% 86.4 87.7 87.7 89.0 87.0 87.0 81.4 78.8 78.8 79.4 79.4 76.7 76.7 18.6 6,931 364 620 124 441 1,009 1,384 133 705 910 1,056 115 350 444 382 1,119 For the exclusive use of M. Gipson, 2020. For article reprints call 800-988-0886 or 617-783-7500, or visit hbr.org Figure 3: Refine the Forecast brand-mileage warranty combination sizes National National HIGH MEDIUM House 1 HIGH House 2 HIGH totals House 2 House 3 Total Share of demand Total MEDIUM LOW units sold captured demand A 100 29 28 190 347 B 282 21 30 203 536 C 11 12 86 109 D 53 50 284 387 E 72 64 20 172 570 898 F 59 97 285 763 1,204 G 10 16 14 76 116 H 7 33 157 377 574 I 10 183 524 717 J 39 225 568 832 K 8 10 73 91 L 8 47 223 278 M 43 298 341 N 72 221 293 O 8 200 208 Total units sold 531 141 182 Share of total 7.7% 2.0% 2.6% sales Best-fit demand 2.4% 1.1% 1.5% shares 1,331 19.2% 3,986 57.5% 760 11.0% 6.7% 18.6% 69.6% adding up the shares of total sales enjoyed by each of the combinations offered in size F (7.7% + 2.6% + 19.2% + 57.5%). That tells us the share of total demand for size F that the retailer is currently capturing (87%). In other words, the retailer is theoretically forfeiting the shares of sales associated with the two combinations it does not offer: National Medium, 2%, and House 3 Low, 11%. To calculate total demand for size F, we simply divide total sales for size F by the share of demand captured: 1,204 ÷ 87% = 1,384. Once we know the total demand for size F, we can estimate demand for any SKU in that size, by multiplying the total demand for the size by the share of sales enjoyed by the brandwarranty combination. For example, House 3 Low has an overall share of 11%; applying that percentage to 1,384 gives us a forecast of 152 units for House 3 Low in size F. Refine the forecast. This calculation takes us only partway to an accurate forecast, as you’ll see if you forecast the sales of an item you actually carry. We determined above that if all combinations for size F were offered, the total sales would be 1,384 units. Thus, the demand for House 2 Medium, the retailer’s top seller, is estimated at 796 (1,384 x 57.5%). Actual sales, however, were quite a bit lower: 763 units. One reason for the discrepancy is that sales shares are influenced by the assortment offered. House 2 High and Medium were offered in almost every size, which raised their total sales shares relative 97.3% 28.8 94.9 94.9 30.3 29.2 29.2 27.9 26.4 26.4 26.8 26.8 25.3 25.3 72.0 357 1,862 115 408 2,960 4,123 397 2,054 2,717 3,153 339 1,037 1,350 1,160 289 6,931 Because estimates have inherent margins of error, the retailer refined the forecasts to produce “best fit” demand shares. Using best-fit shares, demand for size f almost triples: 1,204÷ 29.2%= 4,123 total share demand captured for size F total demand for size F ESTIMATES MAY NOT BE PERFECTLY ACCURATE: 4,123 × 18.6%= 767 total share demand captured for size F for House 2 Medium demand estimate for House 2 Medium size F 4,123 × 6.7% = 276 total share demand captured for size F for House 2 High demand estimate for House 2 High size F to those of brand-warranty combinations offered in fewer sizes. To correct for such discrepancies, we need to tweak the brand-warranty sales shares to minimize the average difference—what statisticians call the mean absolute deviation—between estimates and actual sales. This highly iterative process is done using an optimization tool like Excel Solver. Essentially, the tool plugs trial values for the brand-warranty share numbers into the demand estimate calculations for all current SKUs and sees how close the resultant forecasts are to actual sales. Then it adjusts the share values to make forecasts closer, and repeats until it arrives at the share values that minimize the sum of all discrepancies over the SKUs offered. It’s exactly the way you get a prescription for eyeglasses: Start with a trial lens, try a different lens— November 2012 Harvard Business Review 7 This document is authorized for use only by Mika Gipson in 2020. For the exclusive use of M. Gipson, 2020. Which Products Should You Stock? Figure 4: Estimate How Well New SKUs Would Sell brand-mileage warranty combination National National HIGH MEDIUM House 2 HIGH House 2 House 3 MEDIUM LOW A 4 6 B 29 1,297 C 3 1 2 D 10 5 6 E 2,060 F 46 2,862 G 4 276 H 49 1,430 I 65 43 1,892 J 75 50 2,196 K 8 4 236 L 25 12 721 M 32 15 21 940 N 28 13 18 808 O 3 5 19 54 sizes Using the best-fit shares, the retailer estimated demand for all SKUs it could add to the product mix. House 1 HIGH better or worse?—adjust, and repeat until there’s no further improvement. This process resulted in “best-fit” demand shares for the six brand-warranty combinations: 2.4%, 1.1%, 1.5%, 6.7%, 18.6%, and 69.6%. (See Figure 3.) Compare the demand shares with actual sales shares, and a dramatically different picture of optimal assortment begins to emerge. Note that the forecasts, while close to actual sales, are not perfectly accurate. Two factors contribute to forecast errors: First, there are random fluctuations in sales. And second, our assumption that demand for a SKU equals the demand for the size multiplied by the brand-warranty share is imperfect, because the shares of brand-warranty combinations can differ by size. (For example, demand was higher for low-end tires in sizes that fit older, less expensive cars than in other sizes.) Having determined the best-fit share values, demand can be estimated for all potential SKUs. (See Figure 4.) Account for trading up and down. Now let’s consider another wrinkle: substitution. The calculations we described above do not explicitly account for the fact that customers might be willing to buy a different brand-warranty combination if their preferred option is not offered. For example, the retailer suspected that the 57.5% sales share of House 2 Medium did not necessarily mean that more than half of its customers preferred this combination; they might have settled for it because their preferred option, the cheaper House 3 Low, wasn’t offered. One clue that this could be the case was that when House 3 Low was offered in a 8 Harvard Business Review November 2012 given size, it outsold House 2 Medium by about six to one. Making matters even more complicated, the degree to which customers trade up and down may not be the same for all quality levels. If you think that’s the case for SKUs that are especially important to your business, you need to account for this in your calculations. For the tire project, we assumed that the fractions of customers who would trade up or down were equal for all brand-warranty combinations with the exception of customers shifting from House 3 Low to House 2 Medium. (Those two brandwarranty combinations accounted for more than two-thirds of sales.) So our model now requires nine parameters: the six brand-warranty shares and three substitution parameters, which include the fraction of customers who trade up one quality level, who trade down one quality level, and who shift from House 3 Low to House 2 Medium. As before, we use a tool such as Excel Solver that plugs in trial values for the shares and the fractions, calculates demand estimates and sees how close the resultant forecasts are to actual sales. It adjusts the shares and fractions to make forecasts closer, and repeats until there is no further improvement. The final results: 35% of customers who couldn’t find House 3 Low in the assortment in their size would trade up and buy House 2 Medium. For other quality levels, 2% would trade up and 1% would trade down if they couldn’t find what they were looking for. Once you know the fractions of customers trading up or down, you can account for substitution in your demand estimates. Consider House 2 Medium This document is authorized for use only by Mika Gipson in 2020. For the exclusive use of M. Gipson, 2020. For article reprints call 800-988-0886 or 617-783-7500, or visit hbr.org in size F. Because House 3 Low is not offered in this size, take the demand estimate for House 2 Medium and add to it that of House 3 Low multiplied by the fraction of customers who would trade up. For size A, in contrast, both House 2 Medium and House 3 Low are in the assortment, so no value for substitution demand is figured into House 2 Medium’s estimate. Look for self-fulfilling prophesies. Now consider this familiar scenario: A retailer thinks its customers don’t want to buy a certain product type (or the retailer doesn’t want to carry it). So the company offers a limited amount and thus doesn’t sell much of it, seemingly confirming the original assumption that customers don’t want it. But in the end, the assortment reflects the products the retailer wants to carry rather than those its customers want to buy— a risky proposition. One benefit of our technique is that it allows retailers to spot such situations. For instance, in comparing estimated demand and actual sales at the auto retailer, we found one surprising result: House 3 Low had an 11% sales share, but our estimates pegged demand at a whopping 69.6% of total sales. The low sales share occurred because the retailer offered this very-low-priced tire in only a few sizes and therefore didn’t sell many. But as the data show, when customers had a choice between House 3 Low and House 2 Medium, they strongly preferred the former. This pattern persisted This is not as simple as identifying the top 100 revenue-generating SKUs and calling it an assortment. chainwide. There were nine sizes in which House 3 Low and House 2 Medium were both offered, and in every case, House 3 Low outsold House 2 Medium in total chain sales—by more than seven to one. The retailer offered a limited selection of the cheapest tire because its managers thought they could trade customers up to the somewhat-higherpriced House 2 Medium. But they were successful in upselling only 35% of the time. Indeed, our model shows that by ignoring the estimated 69.6% share of House 3 Low, the company was losing 45% of its potential sales (the 65% of the 69.6% of customers who want House 3 Low and don’t trade up). To gain more insight into this finding, we tabulated average income in the area served by each store and used it to create Figure 5, which shows that the share of the cheapest House 3 Low and unwillingness to trade up were inversely correlated with income. In other words, the lower the average income Figure 5: How Income Affects Demand and Willingness to Trade Up 44% from house 3 low to house 2 medium customers willing to trade up 80% house 3 low share of sales 60 40 20 The lower the average income of the area the tire store served, the more customers preferred the leastexpensive tires... 20–30 30–40 40–50 50–60 60–70 70–80 80–90 90–100 median household income (thousands us$) 40 ...and the less willing they were to trade up to more-expensive tires. 36 32 20–35 35–50 50–75 75–100 median household income (Thousands US$) November 2012 Harvard Business Review 9 This document is authorized for use only by Mika Gipson in 2020. For the exclusive use of M. Gipson, 2020. Which Products Should You Stock? Our approach is superior to the conventional one, which requires retailers to guess how attributes influence demand. of the area that a store serves, the more its customers preferred the least-expensive tires and the less willing they were to trade up to more-expensive ones. Optimize the Assortment We will now describe how to use our model to decide which existing and new SKUs would constitute an optimal assortment. 1. Decide whether to maximize revenues or profits. The most natural profit measure in a retail context is total gross margin—typically, revenues minus cost of goods sold. Business schools and economists preach profit maximization, but retailers also care about revenues, in part because Wall Street watches that metric closely. In both the tire example and in the appearance-chemicals case, which we’ll look at later, the goal was to maximize revenue. 2. Decide on pricing for potential SKUs. To optimize an assortment, you need to know how Figure 6: How Many SKUs Should You Carry? 250 One assortment per store 200 revenue (millions us$) 150 One assortment for the chain 100 50 0 100 200 number of skus in the assortment 10 Harvard Business Review November 2012 300 400 much revenue (or margin) each SKU would generate. Prices are a key input in this calculation. The prices of existing SKUs are known, of course. If prices for the new SKUs are unavailable, come up with estimates by comparing the attributes of current SKUs with those of potential new ones. In the tire example, we observed that the prices of a given size decreased consistently from the highestpriced brand-warranty combination (National High) to the lowest-priced (House 3 Low). We applied those decreases to estimate the prices of SKUs not carried. 3. Decide on the final assortments. Next, calculate the potential revenue of each SKU by multiplying its forecasted unit sales by its retail price. Now you have the data you need to begin building your assortment. Start with the SKU that would generate the greatest revenue or profit for the store or the chain. Then add the SKU that would yield the second-greatest increase in revenue. Continue to add SKUs until you hit the maximum number of SKUs you want to carry, say, 100 out of a possible universe of 400. Make no mistake: This is not as simple as identifying the top 100 revenue-generating SKUs and calling it an assortment. Because of demand substitution, each time you add a SKU to the assortment, you have to adjust your figures to account for how that new SKU affects demand for the ones you’ve already added. The process, obviously, is highly iterative. When we applied this process to create an optimal chainwide assortment of tire SKUs for the auto parts retailer, we found that 47 of the 105 SKUs should be replaced. (Not surprisingly, many of the proposed replacement SKUs were House 3 Low.) The retailer partially implemented our recommendations: It added 10 new SKUs and deleted 10 others. One reason for partial implementation was that the retailer couldn’t find vendors to produce all 47 of the proposed new SKUs. We tracked sales after implementation and found that even this partial adjustment of the assortment increased revenue by 5.8% and gross margin by 4.2%—a significant improvement. Our assortment analysis also prompted the retailer to slightly raise the price of the cheapest tire and lower that of the next-most-expensive tire, increasing the chances that customers would trade up. The beauty of our approach is that it lets you see how revenue varies with the breadth of the assortment. Figure 6 shows how the tire retailer’s revenues This document is authorized for use only by Mika Gipson in 2020. For the exclusive use of M. Gipson, 2020. For article reprints call 800-988-0886 or 617-783-7500, or visit hbr.org are influenced by the number of SKUs in the assortment. The upper line shows the revenues if each store had its optimal assortment, while the lower line gives the revenues for a single optimal assortment for the chain. Graphs like this can be used to change the shelf space allotted to categories in order to increase sales. They also can help retailers avoid the mistakes made by Walmart and Super Fresh, which reduced their assortment breadth with disastrous results. Localizing the Assortment Creating localized assortments is complicated. A retailer needs to understand how demand differs between stores and then create assortments that cater to store-specific tastes. Most retailers find it much too complicated to carry a unique assortment for each store; instead, they create clusters of stores that use the same assortment. In such cases, they need to decide how many clusters to create, on what basis clusters should be formed (for instance, income or weather), and what assortment to use in each cluster. Our attribute-based technique is an excellent way to answer these questions, as we can see in a study of the appearance-chemicals category that we performed for an auto parts retailer that had hundreds of stores. Appearance chemicals comprises an array of liquids and pastes used to wash, wax, shine, polish, and protect cars. In the study, we identified six attributes of products in the category: the car surface to be treated, what is to be done to it, the application mode, the package size, the brand, and the quality level (good, better, or best). The retailer was eager to understand how demand patterns differed across stores and then use that information to localize its assortments. It considered at most five clusters, believing that it would not be operationally feasible to have more than five assortments. We applied the method described above to estimate demand shares for the various attribute levels, used those estimates to forecast demand for potential new SKUs, and generated a revenue-maximizing assortment for each store. Using the assortments we had generated, we then created store clusters. We began the process by assuming that each store represented a cluster. We then identified the two stores that would suffer the smallest reduction in revenues if they were forced to share the same assortment and combined them to create a two-store cluster. We repeated the process, identifying the next two stores that could best share an assortment or by adding a third store to the two we had already combined—whichever would result in the smallest reduction in revenue. We kept going, each time reducing the number of store clusters by one, until we were left with a single cluster of all stores. This gave us revenue numbers for all levels of localization, ranging from a single assortment for the entire chain up to an individual assortment for each store. Figure 7 shows the revenues for five localization options, ranging from one to five store clusters, adjusted to make the revenues of a single assortment 100. As you can see, the returns from adding store clusters diminish, which led the retailer to implement the two-cluster solution. The data also revealed that one of the two clusters, accounting for about a third of the chain’s stores, sold much higher levels of tire-related products. That cluster had a distinctive ethnicity, which the retailer called urban/bilingual. So these stores’ assortments featured more tire-related products, and the retailer created signage that called attention to them. After tracking sales for six months, we found that the chain’s revenue in the appearance-chemicals category was up 3.5%. This gain resulted both from localization and from improving the base assortment. Moreover, the new assortment and new signage helped the retailer in another way: It had been losing sales to the competition in the urban/ bilingual stores, but after the changes, it began to show increases in comparable-store sales for the category. We believe this demand-based approach to clustering is superior to the conventional approach, which requires retailers to guess how store attributes influence demand. Figure 7: How Many Clusters Do You Need? Number of store clusters Revenue 1 2 3 4 5 100.0 102.1 103.5 104.6 105.7 revenue numbers have been indexed The returns from adding store clusters diminishes, which led the retailer to implement a twocluster solution. Analytics have not been heavily applied to assortment planning, especially at the operational level of deciding which SKUs to carry. Our method uses analytics to glean insight into the product attributes your customers prefer at each store and then create localized assortments on the basis of those insights. Assortment planning can add significantly to samestore sales; but done wrong, it can cripple a retailer for years. Our method can help retailers do it well. HBR Reprint R1211J Marshall Fisher is the UPS Professor of Operations and Information Management at the University of Pennsylvania’s Wharton School and a coauthor, with Ananth Raman, of The New Science of Retailing: How Analytics Are Transforming the Supply Chain and Improving Performance (Harvard Business Press, 2010). Ramnath Vaidyanathan is an assistant professor of operations management at McGill University’s Desautels Faculty of Management. November 2012 Harvard Business Review 11 This document is authorized for use only by Mika Gipson in 2020. The Bullwhip Effect in Supply Chains SPRING 1997 Distorted information from one end of a supply chain to the other can lead to tremendous inefficiencies: excessive inventory investment, poor customer service, lost revenues, misguided capacity plans, ineffective transportation, and missed production schedules. How do exaggerated order swings occur? What can companies do to mitigate them? Hau L. Lee V. Padmanabhan Seungjin Whang Vol. 38, No. 3 Reprint #3837 http://mitsmr.com/1phEOiM The Bullwhip Effect in Supply Chains Hau L. Lee • V. Padmanabhan • Seungjin Whang Distorted information from one end of a supply chain to the other can lead to tremendous inefficiencies: excessive inventory investment, poor customer service, lost revenues, misguided capacity plans, ineffective transportation, and missed production schedules. How do exaggerated order swings occur? What can companies do to mitigate them? N ot long ago, logistics executives at Procter & Gamble (P&G) examined the order patterns for one of their best-selling products, Pampers. Its sales at retail stores were fluctuating, but the variabilities were certainly not excessive. However, as they examined the distributors’ orders, the executives were surprised by the degree of variability. When they looked at P&G’s orders of materials to their suppliers, such as 3M, they discovered that the swings were even greater. At first glance, the variabilities did not make sense. While the consumers, in this case, the babies, consumed diapers at a steady rate, the demand order variabilities in the supply chain were amplified as they moved up the supply chain. P&G called this phenomenon the “bullwhip” effect. (In some industries, it is known as the “whiplash” or the “whipsaw” effect.) When Hewlett-Packard (HP) executives examined the sales of one of its printers at a major reseller, they found that there were, as expected, some fluctuations SLOAN MANAGEMENT REVIEW/SPRING 1997 over time. However, when they examined the orders from the reseller, they observed much bigger swings. Also, to their surprise, they discovered that the orders from the printer division to the company’s integrated circuit division had even greater fluctuations. What happens when a supply chain is plagued with a bullwhip effect that distorts its demand information as it is transmitted up the chain? In the past, without being able to see the sales of its products at the distribution channel stage, HP had to rely on the sales orders from the resellers to make product forecasts, plan capacity, control inventory, and schedule production. Big variations in demand were a major problem for HP’s management. The common symptoms of such variations could be excessive inventory, poor product forecasts, insufficient or excessive capacities, poor customer service due to unavailable products or long backlogs, uncertain production planning (i.e., excessive revisions), and high costs for corrections, such as for expedited shipments and overtime. HP’s product division was a victim of order swings that were exaggerated by the resellers relative to their sales; it, in turn, created additional exaggerations of order swings to suppliers. In the past few years, the Efficient Consumer Response (ECR) initiative has tried to redefine how the grocery supply chain should work.1 One motivation for the initiative was the excessive amount of inventory in the supply chain. Various industry studies found that the total supply chain, from when products leave the manufacturers’ production lines to when they arrive on the retailers’ shelves, has more than 100 days of Hau L. Lee is the Kleiner Perkins, Mayfield, Sequoia Capital Professor in Industrial Engineering and Engineering Management, and professor of operations management at the Graduate School of Business, Stanford University. V. Padmanabhan is an associate professor of marketing, and Seungjin Whang is an associate professor of operations information and technology, also at Stanford. LEE ET AL. 93 Figure 1 Increasing Variability of Orders up the Supply Chain Retailer's Orders to Manufacturer 20 20 15 15 Order Quantity Order Quantity Consumer Sales 10 5 0 Time Wholesaler's Orders to Manufacturer Manufacturer's Orders to Supplier 20 20 15 15 10 5 Time inventory supply. Distorted information has led every entity in the supply chain — the plant warehouse, a manufacturer’s shuttle warehouse, a manufacturer’s market warehouse, a distributor’s central warehouse, the distributor’s regional warehouses, and the retail store’s storage space — to stockpile because of the high degree of demand uncertainties and variabili- T he ordering patterns share a common, recurring theme: the variabilities of an upstream site are always greater than those of the downstream site. ties. It’s no wonder that the ECR reports estimated a potential $30 billion opportunity from streamlining the inefficiencies of the grocery supply chain.2 Other industries are in a similar position. Computer factories and manufacturers’ distribution centers, the LEE ET AL. 10 5 0 0 94 5 Time Order Quantity Order Quantity 0 10 Time distributors’ warehouses, and store warehouses along the distribution channel have inventory stockpiles. And in the pharmaceutical industry, there are duplicated inventories in a supply chain of manufacturers such as Eli Lilly or Bristol-Myers Squibb, distributors such as McKesson, and retailers such as Longs Drug Stores. Again, information distortion can cause the total inventory in this supply chain to exceed 100 days of supply. With inventories of raw materials, such as integrated circuits and printed circuit boards in the computer industry and antibodies and vial manufacturing in the pharmaceutical industry, the total chain may contain more than one year’s supply. In a supply chain for a typical consumer product, even when consumer sales do not seem to vary much, there is pronounced variability in the retailers’ orders to the wholesalers (see Figure 1). Orders to the manufacturer and to the manufacturers’ supplier spike even more. To solve the problem of distorted information, companies need to first understand what creates the bullwhip effect so they can counteract it. Innovative companies in different industries have found that they SLOAN MANAGEMENT REVIEW/SPRING 1997 can control the bullwhip effect and improve their supply chain performance by coordinating information and planning along the supply chain. Causes of the Bullwhip Effect Perhaps the best illustration of the bullwhip effect is the well-known “beer game.”3 In the game, participants (students, managers, analysts, and so on) play the roles of customers, retailers, wholesalers, and suppliers of a popular brand of beer. The participants cannot communicate with each other and must make order decisions based only on orders from the next downstream player. The ordering patterns share a common, recurring theme: the variabilities of an upstream site are always greater than those of the downstream site, a simple, yet powerful illustration of the bullwhip effect. This amplified order variability may be attributed to the players’ irrational decision making. Indeed, Sterman’s experiments showed that human behavior, such as misconceptions about inventory and demand information, may cause the bullwhip effect.4 In contrast, we show that the bullwhip effect is a consequence of the players’ rational behavior within the supply chain’s infrastructure. This important distinction implies that companies wanting to control the bullwhip effect have to focus on modifying the chain’s infrastructure and related processes rather than the decision makers’ behavior. We have identified four major causes of the bullwhip effect: 1. Demand forecast updating 2. Order batching 3. Price fluctuation 4. Rationing and shortage gaming Each of the four forces in concert with the chain’s infrastructure and the order managers’ rational decision making create the bullwhip effect. Understanding the causes helps managers design and develop strategies to counter it.5 Demand Forecast Updating Every company in a supply chain usually does product forecasting for its production scheduling, capacity planning, inventory control, and material requirements planning. Forecasting is often based on the order history from the company’s immediate customers. SLOAN MANAGEMENT REVIEW/SPRING 1997 The outcomes of the beer game are the consequence of many behavioral factors, such as the players’ perceptions and mistrust. An important factor is each player’s thought process in projecting the demand pattern based on what he or she observes. When a downstream operation places an order, the upstream manager processes that piece of information as a signal about future product demand. Based on this signal, the upstream manager readjusts his or her demand forecasts and, in turn, the orders placed with the suppliers of the upstream operation. We contend that demand signal processing is a major contributor to the bullwhip effect. For example, if you are a manager who has to determine how much to order from a supplier, you use a simple method to do demand forecasting, such as exponential smoothing. With exponential smoothing, future demands are continuously updated as the new daily demand data become available. The order you send to the supplier reflects the amount you need to replenish the stocks to meet the requirements of future demands, as well as the necessary safety stocks. The future demands and the associated safety stocks are updated using the smoothing technique. With long lead times, it is not uncommon to have weeks of safety stocks. The result is that the fluctuations in the order quantities over time can be much greater than those in the demand data. Now, one site up the supply chain, if you are the manager of the supplier, the daily orders from the manager of the previous site constitute your demand. If you are also using exponential smoothing to update your forecasts and safety stocks, the orders that you place with your supplier will have even bigger swings. For an example of such fluctuations in demand, see Figure 2. As we can see from the figure, the orders placed by the dealer to the manufacturer have much greater variability than the consumer demands. Because the amount of safety stock contributes to the bullwhip effect, it is intuitive that, when the lead times between the resupply of the items along the supply chain are longer, the fluctuation is even more significant. Order Batching In a supply chain, each company places orders with an upstream organization using some inventory monitoring or control. Demands come in, depleting inven- LEE ET AL. 95 Figure 2 Higher Variability in Orders from Dealer to Manufacturer than Actual Sales 60 Quantity 50 40 Orders Placed 30 20 Actual Sales 10 0 Time tory, but the company may not immediately place an order with its supplier. It often batches or accumulates demands before issuing an order. There are two forms of order batching: periodic ordering and push ordering. Instead of ordering frequently, companies may order weekly, biweekly, or even monthly. There are many common reasons for an inventory system based on order cycles. Often the supplier cannot handle frequent order processing because the time and cost of processing an order can be substantial. P&G estimated that, because of the many manual interventions needed in its order, billing, and shipment systems, each invoice to its customers cost between $35 and $75 to process.6 Many manufacturers place purchase orders with suppliers when they run their material requirements planning (MRP) systems. MRP systems are often run monthly, resulting in monthly ordering with suppliers. A company with slow-moving items may prefer to order on a regular cyclical basis because there may not be enough items consumed to warrant resupply if it orders more frequently. Consider a company that orders once a month from its supplier. The supplier faces a highly erratic stream of orders. There is a spike in demand at one time during the month, followed by no demands for the rest of the month. Of course, this variability is higher than the demands the company itself faces. Periodic ordering amplifies variability and contributes to the bullwhip effect. One common obstacle for a company that wants to order frequently is the economics of transportation. There are substantial differences between full truck- 96 LEE ET AL. load (FTL) and less-than-truckload rates, so companies have a strong incentive to fill a truckload when they order materials from a supplier. Sometimes, suppliers give their best pricing for FTL orders. For most items, a full truckload could be a supply of a month or more. Full or close to full truckload ordering would thus lead to moderate to excessively long order cycles. In push ordering, a company experiences regular surges in demand. The company has orders “pushed” on it from customers periodically because salespeople are regularly measured, sometimes quarterly or annually, which causes end-of-quarter or end-of-year order surges. Salespersons who need to fill sales quotas may “borrow” ahead and sign orders prematurely. The U.S. Navy’s study of recruiter productivity found surges in the number of recruits by the recruiters on a periodic cycle that coincided with their evaluation cycle.7 For companies, the ordering pattern from their customers is more erratic than the consumption patterns that their customers experience. The “hockey stick” phenomenon is quite prevalent. When a company faces periodic ordering by its customers, the bullwhip effect results. If all customers’ order cycles were spread out evenly throughout the A lthough some companies claim to thrive on high-low buying practices,most suffer. week, the bullwhip effect would be minimal. The periodic surges in demand by some customers would be insignificant because not all would be ordering at the same time. Unfortunately, such an ideal situation rarely exists. Orders are more likely to be randomly spread out or, worse, to overlap. When order cycles overlap, most customers that order periodically do so at the same time. As a result, the surge in demand is even more pronounced, and the variability from the bullwhip effect is at its highest. If the majority of companies that do MRP or distribution requirement planning (DRP) to generate purchase orders do so at the beginning of the month (or end of the month), order cycles overlap. Periodic SLOAN MANAGEMENT REVIEW/SPRING 1997 Price Fluctuation Estimates indicate that 80 percent of the transactions between manufacturers and distributors in the grocery industry were made in a “forward buy” arrangement in which items were bought in advance of requirements, usually because of a manufacturer’s attractive price offer.8 Forward buying constitutes $75 billion to $100 billion of inventory in the grocery industry.9 Forward buying results from price fluctuations in the marketplace. Manufacturers and distributors periodically have special promotions like price discounts, quantity discounts, coupons, rebates, and so on. All these promotions result in price fluctuations. Additionally, manufacturers offer trade deals (e.g., special discounts, price terms, and payment terms) to the distributors and wholesalers, which are an indirect form of price discounts. For example, Kotler reports that trade deals and consumer promotion constitute 47 percent and 28 percent, respectively, of their total promotion budgets.10 The result is that customers buy in quantities that do not reflect their immediate needs; they buy in bigger quantities and stock up for the future. Such promotions can be costly to the supply chain.11 What happens if forward buying becomes the norm? When a product’s price is low (through direct discount or promotional schemes), a customer buys in bigger quantities than needed. When the product’s price returns to normal, the customer stops buying until it has depleted its inventory. As a result, the customer’s buying pattern does not reflect its consumption pattern, and the variation of the buying quantities is much bigger than the variation of the consumption rate — the bullwhip effect. When high-low pricing occurs, forward buying may well be a rational decision. If the cost of holding inventory is less than the price differential, buying in advance makes sense. In fact, the high-low pricing phenomenon has induced a stream of research on how companies should order optimally to take advantage of the low price opportunities. Although some companies claim to thrive on high-low buying practices, most suffer. For example, a soup manufacturer’s leading brand has seasonal SLOAN MANAGEMENT REVIEW/SPRING 1997 Figure 3 Bullwhip Effect due to Seasonal Sales of Soup Weekly Quantity execution of MRPs contributes to the bullwhip effect, or “MRP jitters” or “DRP jitters.” 800 700 600 500 400 300 200 100 0 Shipments from Manufacturer to Distributors Retailers' Sales 1 52 Weeks sales, with higher sales in the winter (see Figure 3). However, the shipment quantities from the manufacturer to the distributors, reflecting orders from the distributors to the manufacturer, varied more widely. When faced with such wide swings, companies often have to run their factories overtime at certain times and be idle at others. Alternatively, companies may have to build huge piles of inventory to anticipate big swings in demand. With a surge in shipments, they may also have to pay premium freight rates to transport products. Damage also increases from handling larger than normal volumes and stocking inventories for long periods. The irony is that these variations are induced by price fluctuations that the manufacturers and the distributors set up themselves. It’s no wonder that such a practice was called “the dumbest marketing ploy ever.”12 Using trade promotions can backfire because of the impact on the manufacturers’ stock performance. A group of shareholders sued Bristol-Myers Squibb when its stock plummeted from $74 to $67 as a result of a disappointing quarterly sales performance; its actual sales increase was only 5 percent instead of the anticipated 13 percent. The sluggish sales increase was reportedly due to the company’s trade deals in a previous quarter that flooded the distribution channel with forward-buy inventories of its product.13 Rationing and Shortage Gaming When product demand exceeds supply, a manufacturer often rations its product to customers. In one scheme, the manufacturer allocates the amount in proportion to the amount ordered. For example, if the total supply is only 50 percent of the total demand, all customers LEE ET AL. 97 receive 50 percent of what they order. Knowing that the manufacturer will ration when the product is in short supply, customers exaggerate their real needs when they order. Later, when demand cools, orders will suddenly disappear and cancellations pour in. This seeming overreaction by customers anticipating shortages results when organizations and individuals make sound, rational economic decisions and “game” the potential rationing.14 The effect of “gaming” is that customers’ orders give the supplier little information on the product’s real demand, a particularly vexing problem for manufacturers in a product’s early stages. The gaming practice is very common. In the 1980s, on several occasions, the computer industry perceived a shortage of DRAM chips. Orders shot up, not because of an increase in consumption, but because of anticipation. Customers place duplicate orders with multiple suppliers and buy from the first one that can deliver, then cancel all other duplicate orders.15 More recently, Hewlett-Packard could not meet the demand for its LaserJet III printer and rationed the product. Orders surged, but HP managers could not discern whether the orders genuinely reflected real market demands or were simply phantom orders from resellers trying to get better allocation of the product. When HP lifted its constraints on resupply of the LaserJets, many resellers canceled their orders. HP’s costs in excess inventory after the allocation period and in unnecessary capacity increases were in the millions of dollars.16 During the Christmas shopping seasons in 1992 and 1993, Motorola could not meet consumer demand for handsets and cellular phones, forcing many distributors to turn away business. Distributors like AirTouch Communications and the Baby Bells, anticipating the possibility of shortages and acting defensively, drastically overordered toward the end of 1994.17 Because of such overzealous ordering by retail distributors, Motorola reported record fourth-quarter earnings in January 1995. Once Wall Street realized that the dealers were swamped with inventory and new orders for phones were not as healthy before, Motorola’s stock tumbled almost 10 percent. In October 1994, IBM’s new Aptiva personal computer was selling extremely well, leading resellers to speculate that IBM might run out of the product before the Christmas season. According to some analysts, 98 LEE ET AL. IBM, hampered by an overstock problem the previous year, planned production too conservatively. Other analysts referred to the possibility of rationing: “Retailers — apparently convinced Aptiva will sell well and afraid of being left with insufficient stock to meet holiday season demand — increased their orders with IBM, believing they wouldn’t get all they asked for.”18 It was unclear to IBM how much of the increase in orders was genuine market demand and how much was due to resellers placing phantom orders when IBM had to ration the product. How to Counteract the Bullwhip Effect Understanding the causes of the bullwhip effect can help managers find strategies to mitigate it. Indeed, many companies have begun to implement innovative programs that partially address the effect. Next we examine how companies tackle each of the four causes. We categorize the various initiatives and other possible remedies based on the underlying coordination mechanism, namely, information sharing, channel alignment, and operational efficiency. With information sharing, demand information at a downstream site is transmitted upstream in a timely fashion. Channel alignment is the coordination of pricing, transportation, inventory planning, and ownership between the upstream and downstream sites in a supply chain. Operational efficiency refers to activities that improve performance, such as reduced costs and lead time. We use this topology to discuss ways to control the bullwhip effect (see Table 1). Avoid Multiple Demand Forecast Updates Ordinarily, every member of a supply chain conducts some sort of forecasting in connection with its planning (e.g., the manufacturer does the production planning, the wholesaler, the logistics planning, and so on). Bullwhip effects are created when supply chain members process the demand input from their immediate downstream member in producing their own forecasts. Demand input from the immediate downstream member, of course, results from that member’s forecasting, with input from its own downstream member. One remedy to the repetitive processing of consumption data in a supply chain is to make demand data at a downstream site available to the upstream site. Hence, SLOAN MANAGEMENT REVIEW/SPRING 1997 both sites can update their forecasts Table 1 A Framework for Supply Chain Coordination Initiatives with the same raw data. In the computer industry, manufacturers request Causes of Information Channel Operational Bullwhip Sharing Alignment Efficiency sell-through data on withdrawn stocks from their resellers’ central warehouse. Demand • Understanding • Vendor-managed • Lead-time reduction Although the data are not as complete Forecast system dynamics inventory (VMI) • Echelon-based as point-of-sale (POS) data from the Update • Use point-of-sale • Discount for inforinventory control (POS) data mation sharing resellers’ stores, they offer significantly • Electronic data • Consumer direct more information than was available interchange (EDI) when manufacturers didn’t know what • Internet • Computer-assisted happened after they shipped their ordering (CAO) products. IBM, HP, and Apple all reOrder • EDI • Discount for truck• Reduction in fixed quire sell-through data as part of their Batching • Internet ordering load assortment cost of ordering by contract with resellers. • Delivery appointEDI or electronic Supply chain partners can use elecments commerce • Consolidation • CAO tronic data interchange (EDI) to share • Logistics outdata. In the consumer products indussourcing try, 20 percent of orders by retailers of Price • Continuous • Everyday low price consumer products was transmitted Fluctuations replenishment (EDLP) via EDI in 1990.19 In 1992, that figprogram (CRP) • Activity-based • Everyday low cost costing (ABC) ure was close to 40 percent and, in (EDLC) 1995, nearly 60 percent. The increasShortage • Sharing sales, • Allocation based ing use of EDI will undoubtedly faGaming capacity, and on past sales cilitate information transmission and inventory data sharing among chain members. Even if the multiple organizations in a supply chain use the same source demand data to tor, companies such as Texas Instruments, HP, Motorola, perform forecast updates, the differences in forecasting and Apple use VMI with some of their suppliers and, in methods and buying practices can still lead to unnec- some cases, with their customers. essary fluctuations in the order data placed with the Inventory researchers have long recognized that upstream site. In a more radical approach, the upmulti-echelon inventory systems can operate better stream site could control resupply from upstream to when inventory and demand information from downdownstream. The upstream site would have access to stream sites is available upstream. Echelon inventory the demand and inventory information at the down- — the total inventory at its upstream and downstream stream site and update the necessary forecasts and resites — is key to optimal inventory control.20 Another approach is to try to get demand informasupply for the downstream site. The downstream site, tion about the downstream site by bypassing it. Apple in turn, would become a passive partner in the supply Computer has a “consumer direct” program, i.e., it chain. For example, in the consumer products industry, this practice is known as vendor-managed inven- sells directly to consumers without going through the reseller and distribution channel. A benefit of the protory (VMI) or a continuous replenishment program gram is that it allows Apple to see the demand patterns (CRP). Many companies such as Campbell Soup, for its products. Dell Computers also sells its products M&M/Mars, Nestlé, Quaker Oats, Nabisco, P&G, directly to consumers without going through the disand Scott Paper use CRP with some or most of their tribution channel. customers. Inventory reductions of up to 25 percent are Finally, as we noted before, long resupply lead times common in these alliances. P&G uses VMI in its diacan aggravate the bullwhip effect. Improvements in per supply chain, starting with its supplier, 3M, and its customer, Wal-Mart. Even in the high-technology sec- operational efficiency can help reduce the highly vari- SLOAN MANAGEMENT REVIEW/SPRING 1997 LEE ET AL. 99 able demand due to multiple forecast updates. Hence, just-in-time replenishment is an effective way to mitigate the effect. Break Order Batches Since order batching contributes to the bullwhip effect, companies need to devise strategies that lead to smaller batches or more frequent resupply. In addition, the counterstrategies we described earlier are useful. When an upstream company receives consumption data on a fixed, periodic schedule from its downstream customers, it will not be surprised by an unusually large batched order when there is a demand surge. One reason that order batches are large or order frequencies low is the relatively high cost of placing an order and replenishing it. EDI can reduce the cost of the paperwork in generating an order. Using EDI, companies such as Nabisco perform paperless, computer-assisted ordering (CAO), and, consequently, customers order more frequently. McKesson’s Economost ordering system uses EDI to lower the transaction costs from orders by drugstores and other retailers.21 P&G has introduced standardized ordering terms across all business units to simplify the process and dramatically cut the number of invoices.22 And General Electric is electronically matching buyers and suppliers throughout the company. It expects to purchase at least $1 billion in materials through its internally developed Trading Process Network. A paper purchase order that typically cost $50 to process is now $5.23 Another reason for large order batches is the cost of transportation. The differences in the costs of full truckloads and less-than-truckloads are so great that companies find it economical to order full truckloads, even though this leads to infrequent replenishments from the supplier. In fact, even if orders are made with little effort and low cost through EDI, the improvements in order efficiency are wasted due to the fulltruckload constraint. Now some manufacturers induce their distributors to order assortments of different products. Hence a truckload may contain different products from the same manufacturer (either a plant warehouse site or a manufacturer’s market warehouse) instead of a full load of the same product. The effect is that, for each product, the order frequency is much higher, the frequency of deliveries to the distributors remains unchanged, and the transportation efficiency 100 LEE ET AL. is preserved. P&G has given discounts to distributors that are willing to order mixed-SKU (stock-keeping unit) loads of any of its products.24 Manufacturers could also prepare and ship mixed SKUs to the distributors’ warehouses that are ready to deliver to the stores. “Composite distribution” for fresh produce and chilled products uses the same mixed-SKU concept to make resupply more frequent. Since fresh produce and chilled foods need to be stored at different temperatures, trucks to transport them need to have various temperatures. British retailers like Tesco and Sainsbury use trucks with separate compartments at different temperatures so that they can transport many products on the same truck.25 The use of third-party logistics companies also helps make small batch replenishments economical.26 These companies allow economies of scale that were not feasible in a single supplier-customer relationship. By consolidating loads from multiple suppliers located near each other, a company can realize full truckload economies without the batches coming from the same supplier. Of course, there are additional handling and T he simplest way to control the bullwhip effect caused by forward buying and diversions is to reduce both the frequency and the level of wholesale price discounting. administrative costs for such consolidations or multiple pickups, but the savings often outweigh the costs. Similarly, a third-party logistics company can utilize a truckload to deliver to customers who may be competitors, such as neighboring supermarkets. If each customer is supplied separately via full truckloads, using third-party logistics companies can mean moving from weekly to daily replenishments. For small customers whose volumes do not justify frequent full truckload replenishments independently, this is especially appealing. Some grocery wholesalers that receive FTL shipments from manufacturers and then ship mixed loads to wholesalers’ independent stores use lo- SLOAN MANAGEMENT REVIEW/SPRING 1997 gistics companies. In the United Kingdom, Sainsbury and Tesco have long used National Freight Company for logistics. As a result of the heightened awareness due to the ECR initiative in the grocery industry, we expect to see third-party logistics companies that forecast orders, transport goods, and replenish stores with mixed-SKU pallets from the manufacturers. When customers spread their periodic orders or replenishments evenly over time, they can reduce the negative effect of batching. Some manufacturers coordinate their resupply with their customers. For example, P&G coordinates regular delivery appointments with its customers. Hence, it spreads the replenishments to all the retailers evenly over a week. Stabilize Prices The simplest way to control the bullwhip effect caused by forward buying and diversions is to reduce both the frequency and the level of wholesale price discounting. The manufacturer can reduce the incentives for retail forward buying by establishing a uniform wholesale pricing policy. In the grocery industry, major manufacturers such as P&G, Kraft, and Pillsbury have moved to an everyday low price (EDLP) or value pricing strategy. During the past three years, P&G has reduced its list prices by 12 percent to 24 percent and aggressively slashed the promotions it offers to trade customers. In 1994, P&G reported its highest profit margins in twentyone years and showed increases in market share.27 Similarly, retailers and distributors can aggressively negotiate with their suppliers to give them everyday low cost (EDLC). From 1991 to 1994, the percentage of trade deals in the total promotion budget of grocery products dropped from 50 percent to 47 percent. From an operational perspective, practices such as CRP together with a rationalized wholesale pricing policy can help to control retailers’ tactics, such as diversion. Manufacturers’ use of CAO for sending orders also minimizes the possibility of such a practice. Activity-based costing (ABC) systems enable companies to recognize the excessive costs of forward buying and diversions. When companies run regional promotions, some retailers buy in bulk in the area where the promotions are held, then divert the products to other regions for consumption. The costs of such practices are huge but may not show up in conventional accounting systems. ABC systems provide SLOAN MANAGEMENT REVIEW/SPRING 1997 explicit accounting of the costs of inventory, storage, special handling, premium transportation, and so on that previously were hidden and often outweigh the benefits of promotions. ABC therefore helps companies implement the EDLP strategy.28 Eliminate Gaming in Shortage Situations When a supplier faces a shortage, instead of allocating products based on orders, it can allocate in proportion to past sales records. Customers then have no incentive to exaggerate their orders. General Motors has long used this method of allocation in cases of short supply, and other companies, such as Texas Instruments and Hewlett-Packard, are switching to it. “Gaming” during shortages peaks when customers have little information on the manufacturers’ supply situation. The sharing of capacity and inventory information helps to alleviate customers’ anxiety and, consequently, lessen their need to engage in gaming. But sharing capacity information is insufficient when there is a genuine shortage. Some manufacturers work with customers to place orders well in advance of the sales season. Thus they can adjust production capacity or scheduling with better knowledge of product demand. Finally, the generous return policies that manufacturers offer retailers aggravate gaming. Without a penalty, retailers will continue to exaggerate their needs and cancel orders. Not surprisingly, some computer manufacturers are beginning to enforce more stringent cancellation policies. 2 We contend that the bullwhip effect results from rational decision making by members in the supply chain. Companies can effectively counteract the effect by thoroughly understanding its underlying causes. Industry leaders like Procter & Gamble are implementing innovative strategies that pose new challenges: integrating new information systems, defining new organizational relationships, and implementing new incentive and measurement systems. The choice for companies is clear: either let the bullwhip effect paralyze you or find a way to conquer it. ? References 1. This initiative was engineered by Kurt Salmon Associates but pro- LEE ET AL. 101 pelled by executives from a group of innovative companies like Procter & Gamble and Campbell Soup Company. See: Kurt Salmon Associates, “ECR: Enhancing Consumer Value in the Grocery Industry (Washington, D.C.: report, January 1993); and F.A. Crawford, “ECR: A Mandate for Food Manufacturers?” Food Processing, volume 55, February 1994, pp. 34-42. 2. J.A. Cooke, “The $30 Billion Promise,” Traffic Management, volume 32, December 1993, pp. 57-59. 3. J. Sterman, “Modeling Managerial Behavior: Misperception of Feedback in a Dynamic Decision-Making Experiment,” Management Science, volume 35, number 3, 1989, pp. 321-339. 4. Sterman (1989); and P. Senge, The Fifth Discipline: The Art and Practice of the Learning Organization (New York: Doubleday/Currency, 1990). 5. For a theoretical treatment of this subject, see: H.L. Lee, P. Padmanabhan, and S. Whang, “Information Distortion in a Supply Chain: The Bullwhip Effect,” Management Science, 1997, forthcoming. 6. M. Millstein, “P&G to Restructure Logistics and Pricing,” Supermarket News, 27 June 1994, pp. 1, 49. 7. V. Carroll, H.L. Lee, and A.G. Rao, “Implications of Salesforce Productivity, Heterogeneity and Demotivation: A Navy Recruiter Case Study,” Management Science, volume 32, number 11, 1986, pp. 13711388. 8. Salmon (1993). 9. P. Sellers, “The Dumbest Marketing Ploy,” Fortune, volume 126, 5 October 1992, pp. 88-93. 10. P. Kotler, Marketing Management: Analysis, Planning, Implementation, and Control (Englewood Cliffs, New Jersey: Prentice Hall, 1997). 11. R.D. Buzzell, J.A. Quelch, and W.J. Salmon, “The Costly Bargain of Trade Promotion,” Harvard Business Review, volume 68, MarchApril 1990, pp. 141-148. 12. Sellers (1992). 13. Ibid. 14. Lee et al. (1997). 15. L. Lode, “The Role of Inventory in Delivery Time Competition,” Management Science, volume 38, number 2, 1992, pp. 182-197. 16. Personal communication with Hewlett-Packard. 17. K. Kelly, “Burned by Busy Signals: Why Motorola Ramped up Production Way Past Demand,” Business Week, 6 March 1995, p. 36. 18. Rory J. O’Connor, “Rumor Bolsters IBM Shares,” San Jose Mercury News, 8 October 1994, p. 9D. 19. M. Reid, “Change at the Check-Out,” The Economist, volume 334, 4 March 1995, pp. 3-18. 20. A. Clark and H. Scarf, “Optimal Policies for a Multi-Echelon Inventory Problem,” Management Science, volume 6, number 4, 1960, pp. 465-490. 21. E.K. Clemons and M. Row, “McKesson Drug Company — A Strategic Information System,” Journal of Management Information Systems, volume 5, Summer 1988, pp. 36-50. 22. Millstein (1994). 23. T. Smart, “Jack Welch’s Cyber-Czar,” Business Week, 5 August 1996, pp. 82-83. 24. G. Stern, “Retailers of P&G to Get New Plan on Bills, Shipment,” Wall Street Journal, 22 June 1994. 25. Reid (1995). 26. H.L. Richardson, “How Much Should You Outsource?,” Transportation and Distribution, volume 35, September 1994, pp. 61-62. 27. Z. Schiller, “Ed Artzt’s Elbow Grease Has P&G Shining,” Business Week, 10 October 1994, pp. 84-86. 28. R. Mathews, “CRP Moves Towards Reality,” Progressive Grocer, volume 73, July 1994, pp. 43-44. Reprint 3837 Copyright © 1997 by the SLOAN MANAGEMENT REVIEW ASSOCIATION. All rights reserved. 102 LEE ET AL. SLOAN MANAGEMENT REVIEW/SPRING 1997 MIT SL SLO OAN MANA MANAGEMEN GEMENT T REVIEW OPERA OPERATIONS TIONS PDFs Reprints Permission to Copy Back Issues Articles published in MIT Sloan Management Review are copyrighted by the Massachusetts Institute of Technology unless otherwise specified at the end of an article. MIT Sloan Management Review articles, permissions, and back issues can be purchased on our Web site: sloanreview.mit.edu or you may order through our Business Service Center (9 a.m.-5 p.m. ET) at the phone numbers listed below. Paper reprints are available in quantities of 250 or more. To reproduce or transmit one or more MIT Sloan Management Review articles by electronic or mechanical means (including photocopying or archiving in any information storage or retrieval system) requires written permission. To request permission, use our Web site: sloanreview.mit.edu or E-mail: smr-help@mit.edu Call (US and International):617-253-7170 Fax: 617-258-9739 Posting of full-text SMR articles on publicly accessible Internet sites is prohibited. To obtain permission to post articles on secure and/or passwordprotected intranet sites, e-mail your request to smr-help@mit.edu. Copyright © Massachusetts Institute of Technology, 1997. All rights reserved. Reprint #3837 http://mitsmr.com/1phEOiM Limited Edition Chanel Purse We are going to be creating a new, limited-edition purse inspired by iconic artist Andy Warhol. In collaboration with The Andy Warhol Foundation for the Visual Arts, we will be creating a purse replicating his “Stamped Lips” art piece. Fashion and art are inseparable. They both have the same source of life-creativity. From Dali to Frida, the fashion industry has never stopped paying tribute to art. Thirty years ago, Andy Warhol passed away due to a medical accident. He is regarded as one of the most influential figures in contemporary art and culture, and this influence is not just Campbell cans and Brillo boxes. At the level of philosophies, he broke the boundaries between high and low culture. The playful piece we create explores beauty and identity while also being abstract with its warm and inviting pattern of curving lip lines. The new Stamped Lips purse will be shaped like the lips in his piece and made with red leather and Chanel’s signature diamond stitching. The purse will have the iconic chain with red leather and silver-tone metal interlaced. The classic and timeless CC clasp will seal the Chanel. This limited edition purse is unique in the market because it is the first time the brand collaborates with an artist to produce an item. There will be only 30 purses produced to be distributed worldwide. Proceeds from the Stamped Lips purse sales will be donated to the Andy Warhol Foundation for the Visual Arts, whose mission is to advance the visual arts. About Chanel The company our group selected was Chanel. We chose Chanel because it is easily recognized and considered one of the most popular brands globally, known for its classic and timeless design. Chanel's products are made with the best quality materials and created by the best craftsmen in the industry. Additionally, the brand has a rich history, originating from Coco Chanel, who loved practical yet straightforward fashion pieces. All Chanel items are designed to last and are meant to be worn through decades. Chanel's brand identity reflects Coco Chanel, and being "the ultimate house of luxury, defining style and creating desire, now and forever" (Farfan 2015). Chanel is a symbol of uniqueness, and the brand's personality reflects creativity, which is fitted with our design concept of" unique. " Reference Farfan, B. (2015) What are Luxury Retail Brands Company/Corporate Mission Statements? Availableat: http://retailindustry.about.com/od/CompanyMissionStatementsRetail/fl/What-areLuxury-Retail-Brandsrsquo-CompanyCorporate-Mission-Statements.htm (Accessed: 20th April 2018) Final Project Outline Company X would like to create a new product for the upcoming season. Company X has hired you as Logistics Consultants to create a comprehensive logistics strategy and implementation plan for X product from design to final delivery to the consumer. Each group will create a formal write up and present their completed proposal. Ideas of companies to choose from: 1. Louis Vuitton 2. Gucci 3. Prada 4. Chanel 5. Burberry 6. Coach 7. Michael Kors 8. Patagonia 9. Timberland 10. The North Face 11. Adidas 12. Nike 13. Reebok 14. Asics 15. Lululemon 16. Under Armor 17. Gap 18. Hanes 19. Ralph Lauren 20. Brooks Brothers Formatting Requirements: • Size 12 Times New Roman or Arial Font • Table of Contents • Single spaced • Page numbers (bottom right) • 1-inch margins • Maximum of 12 pages of single-spaced content for parts 3-6 o This does NOT include: Cover Sheet, Table of Contents, Exhibits, Work cited Overall Format: 1. Cover Sheet 2. Table of Contents 3. Executive Summary a. Company Background b. Product Identification (including target customer and current market climate) c. Objectives 4. Strategy Formation a. Supply Chain Overview b. Planning and Forecasting c. Technology and Systems d. Production and Sourcing e. Transportation and Distribution f. Warehousing and Reverse Logistics g. Social Responsibility h. Customer Service 5. Implementation Plan a. Addressing when and how will strategy be executed 6. Conclusion a. Final Recommendations b. Opportunities, implications, market projections. 7. Exhibits (if necessary)