Technology: Supply Chain

By Mel Duvall  |  Posted 05-15-2002

Technology: Supply Chain

Over the past several years, Mike Webb has helped build a global supply chain for Singapore-based Flextronics International Ltd., one of the world's largest makers of electronic components, that's the envy of manufacturers around the world. From their desktops, the company's purchasers, who buy components that will eventually go into products sold by such customers as Nokia Corp., Xerox Corp., Ericsson Inc. and Cisco Systems Inc., can see data pulled together from practically every Flextronics plant on four continents. They know exactly how many components for a given product are available in-house, where they can source more and at what quantity, and which supplier is offering the best price.

But while real advances have been made in improving communication and wringing out inefficiencies, Webb, Flextronics' senior vice president of information technology, is still struggling to overcome the thorniest problem facing his supply chain: accurately forecasting demand. Says Webb: "We take a view that the whole supply chain is a dynamic environment you're constantly trying to keep on top of. But what's ultimately driving all that plumbing is demand. If you can get that part right, the rest is relatively easy."

The fact is, says Webb, you can have the best supply chain technology and the best intelligence-gathering systems available, but they mean little if the information going into the system is flawed, or if the forecasters fail to listen to data they may not want to believe. As John Fontanella, of AMR Research Inc., a Boston technology research firm, points out: "Who's going to tell [Cisco CEO] John Chambers, after he's just had his best year ever, that he'd better cut back production by 30 percent? How would you like to be that sales manager?" And yet, that's exactly what Chambers should have heard much earlier than he did. "There are a lot of tools out there, and they're pretty damn good tools, but they're all dependent on getting good information, and getting that information shared," says Webb. "So there's a cultural issue that's probably just as big as the technological issue."

Making Predictions

Making Predictions

Stanford Professor Hau Lee, who heads up the Stanford Global Supply Chain Management Forum, says manufacturers have been trying to solve demand forecasting for as long as there have been factories. Newer technologies, particularly Internet-based collaboration systems that allow manufacturers, suppliers and customers to work together on new products and jointly form market forecasts, have gone a long way toward improving the accuracy of demand planning. However, Lee says, many other business factors, cultural issues and the inevitable unexpected event will always make forecasting a daunting job. "The technologies we have today help us do a much better job than we ever could before, but sometimes the wheel still falls off the wagon," he says.

Lee points to two critical areas that have plagued companies over the years. First, much of the data people use to develop their forecasts is "polluted," contaminated by special events that happen in the marketplace that can't be easily recognized, or that are utterly unprecedented, such as the events of Sept. 11. But while technology can't tell what is going on in the mind of the customer, it can recognize anomalies in sales patterns, which in turn should cause manufacturers to investigate further before ramping up production. Second is the lack of a feedback mechanism. "People make forecasts, and they know their forecasts will always be wrong to a certain degree, but they don't have the means to learn from it," he says.

Listening to the Data

Listening to the Data

That was the problem at The Bombay Company Inc., a national retail chain that specializes in classic furniture and accessories. Bombay had a nagging 35 percent error rate on its forecasting, including overstocking as well as lost sales due to understocking. Unfortunately, says Roger Tyler, vice president of merchandise planning at the Fort Worth, Texas-based company, a 35 percent error rate is typical for his industry.

Without consistent data on stocking levels for Bombay's more than 400 retail outlets, demand could never be projected accurately. The company was carrying more than $100 million in average inventory at cost—far more than it wanted—and had a 2.02 inventory turnover rate, effectively selling through its inventory only about twice a year. And the company's gross margin return on investment was only 2.61: For every dollar spent on a product stocked in stores, the company was getting $2.61 back—not exactly stellar figures. "The problem as I saw it was that we were using a lot of art in our forecasting, and a little bit of science," says Tyler. "I wanted to be able to use a lot more science, and very little art."

The company went live with a demand planning application from Nonstop Solutions Inc., called Score, in June 2001. Loaded with two years' worth of historical sales information, the system was used to crunch numbers on a wide range of factors, such as the most profitable products, items taking the most shelf space, and the best-selling items in each store. The software allowed Tyler and his team to set service levels for each of the products stocked in stores. Lamps, for example, which sell fast and are highly profitable, could be positioned in the "A" category—a 98 percent "service level"—meaning the product would be made available 98 percent of the time. The software then tracks forecasting errors by flagging discrepancies in service levels and recognizing when an item falls below the ideal level. Meanwhile, administrators can always bump products to higher levels.

By January, the company had reduced its average inventory at cost to $79 million while improving its inventory turnover rate to 2.42. But perhaps the figure Tyler is most happy with is the company's gross margin return on investment, which by January had increased to $3.26—a 25 percent improvement. And the company's overall forecasting error rate, which can now be tracked by the software, has been reduced from 35 percent to about 12 percent. Tyler thinks that figure will be further trimmed as the analytical engine has more sales data to crunch.

Driving on Demand

Driving on Demand

Historical data is useful, but it can't provide the whole picture if a forecasting process dependent upon a reseller channel is fundamentally broken. When Greg O'Neill was parachuted into senior management at Mitsubishi Motor Sales of America Inc. in 1997, he soon realized he had a lemon on his hands.

At the time, Mitsubishi was selling fewer than 190,000 vehicles a year in the U.S. Yet at any given point, the company had close to 45,000 cars parked in inventory at docks, and another 50,000 or so parked on dealer lots—all told, close to half a year's supply. Dealers requested vehicles based on a combination of forecasts provided by the company and their own sense of the local market. But rather than helping to solve the bloated inventory problem, dealers spent much of their time trying to get deep discounts on vehicles they knew Mitsubishi had in excess stock. And Mitsubishi America sales executives at headquarters in Cypress, Calif., only contributed to the fractured process, says O'Neill, the company's executive vice president and general manager. "There was no forecasting going on, because salespeople were too busy trying to figure out ways to wholesale cars to dealers," says O'Neill. "Reacting to market conditions wasn't even in our frame of mind."

Mitsubishi executives decided they would have to draw a line in the sand. "Overnight, we had to say, 'We will never order another car again,'" says O'Neill. "The only order that will come in will have to come from dealers or customers. Our mantra became: The right car, right place, right time." The company began providing dealers with all of the forecasting information and resources that its own corporate forecasters had access to, such as economic data and industry analyst sales forecasts. And it presented dealers with a suggested purchasing plan using numbers crunched by a demand planning tool from supply chain specialist Manugistics Group Inc. The planning tool combines such factors as historical sales trends, expected sales increases from the launch of new products, and forecasted regional growth. The dealers take that purchasing plan and fine-tune it based on their own instincts about their local markets. In the end, the decision on what to buy—and the ultimate responsibility for any errors as a result of those orders—is placed in the dealer's hands.

That may sound like Mitsubishi has abdicated responsibility for forecasting. But the dealers aren't complaining. "They put control where it belonged," says Mike Graeber, owner of Mitsubishi dealerships in San Bernardino, Calif., and the Temecula valley near Los Angeles. "I would much rather forecast my own sales based on what I think will sell than have them forecast my sales based on what they've got in inventory."

Calculating Return

Calculating Return

The results have been impressive. Since 1998, Mitsubishi has reduced its port stock from 45,000 units to none, saving $100 million a year. Cars still come into port and are stored for a short period of time, but Mitsubishi no longer owns any of the vehicles coming in; the dealers do. Meanwhile, cars spend an average of 60 days on dealers lots, down from 166—a significant improvement considering that each car costs Mitsubishi $6 per day in inventory holding costs, the amount a car costs to sit on the lot. Meanwhile, warranty costs have also been reduced, since the longer a vehicle sits on a lot, the more likely it is to pick up nicks and scratches.

The result: Mitsubishi's cost of incentives to dealers—special pricing and no-cost extras thrown into vehicles to generate sales—has been reduced from $2,600 per vehicle in 1998 to $1,700, saving another $290 million a year. Most of that has been redirected into advertising and brand awareness campaigns. O'Neill says Mitsubishi purchased the equivalent of about one week of network television advertising in 1997, and today it purchases 36 weeks. Yet perhaps the most telling figure is that Mitsubishi managed to rev its sales up 69 percent in just three years, from a bit more than 190,000 vehicles in 1998 to more than 320,000 in 2001.

Automotive analyst Dennis DesRosiers, president of DesRosiers Automotive Consultants Inc. of Ontario, Canada, credits moving responsibility for forecasting in- to the hands of dealers as one part of the remarkable turnaround story at Mitsubishi. But he also believes that coming out with the right products at the right time was just as important. In fact, DesRosiers cautions that this kind of "microforecasting" tends to work well in an up cycle, but not as well on the way down. "You've got to do bottom-up as well as top-down forecasting," warns DesRosiers, "or else you're going to get caught when the market swings."

O'Neill disagrees with DesRosiers' assessment. He maintains that Mitsubishi always had a great product. The real difference was in getting the right vehicle to the right place, and in redirecting dollars spent previously on incentives toward marketing. And O'Neill isn't as concerned about "microforecasting," saying that the company's demand planning tool already has the macroeconomic forecasts plugged in, so the dealers are getting the "macro" picture. And he still believes the dealers know their market best. In the first 60 days of this year, he notes, North American auto sales for all manufacturers were down 5 percent from a year earlier, but Mitsubishi America's sales are up 26 percent. "Our dealers chose to ignore some of those macro forecasts—and it's a good thing," he says.

Back on the dealer lot, Graeber says Mitsubishi had better not consider returning to the old methods of forecasting. Now that he has control over his own forecasts, he's not giving it up. "It's about accountability. No one should know my market better than me," says Graeber. "In my opinion, we're a much more customer-driven company than we were in the past."

People in the Data

People in the Data

Yet even being more customer-driven doesn't mean that people in the supply chain can't still foul up the forecasting process. As more companies rely on computers to crunch numbers and automatically spit out orders, common sense can get pushed to the background. In the view of Matt Porta, head of the collaborative value chain practice for PricewaterhouseCoopers, much of last year's oversupply in the telecommunications industry—Solectron Corp. alone built up $4.7 billion in excess inventory—could have been avoided with a little common sense. Porta notes that Solectron was faced with a situation where key customers such as Cisco Systems, Motorola Inc. and Ericsson were all forecasting growth rates of 40 percent or better. If Solectron had added all of the orders up, they would have known something was wrong: The market simply could not grow at that rate. "There is always going to be a certain amount of phantom demand," says Porta. "You have to be able to apply common sense to the picture."

"There are a lot of tools out there, and they're pretty damn good tools, but they're all dependent on getting good information, and getting that information shared," says Flextronics' Webb. "As always, the human element can't be ignored." Webb also believes it's ultimately essential to bring customers into the process to reduce the challenges that people can bring to the process. "The more comfortable we can make our customers in sharing their information and the closer we can get to initial demand for their products," says Webb, "the more accurate we can get at this."


MEL DUVALL is a Calgary-based freelance writer who has covered the technology and business scenes for more than 15 years. Please send comments on this story to editors@cioinsight.com.

Making Predictions

Forecasts By Design

Forecasts by Design thumbnail

Better forecasts come from increased visibility—looking as deeply as possible into both the supply and demand sides of the equation. The "perfect" forecasting process would feed up-to-the-minute data from demand-side sources, and then take into account such external factors as the competition and the impact of major events, as well as internal factors like product lifecycle plans and marketing. But some degree of subjectivity is essential to ensure that the data's decisions make sense.

SOURCE: DR. HAU LEE; CIO INSIGHT

Fact Sheet

Fact Sheet

Strategic Arch. Fachsheet Thumbnail
Click to download a printable version of the Supply Chain Forecasting Fact Sheet (Adobe Acrobat format).

Taking the guesswork out of forecasting is no easy task. But collaborative systems that allow manufacturers, buyers and suppliers to formulate forecasts jointly by developing systems that automatically generate orders on demand can help. Even so, accurate forecasting remains a subjective activity and a problem for IT managers.

Supply Chain Management Forecasting Technology Lets Companies…

  • Evaluate historical data to identify trends and predict demand
  • Collaborate with suppliers and buyers to formulate demand forecasts
  • Automatically create new orders based on real-time sales information
  • Decrease inventory and warehouse costs by reducing inventory hoarding
  • Track and evaluate forecasting errors to learn from mistakes
  • Adapt pricing to move excess inventory and optimize revenue

Advantages

  • Improved accuracy: By getting closer to the source of demand, manufacturers can eliminate phantom orders.
  • Collaboration: Systems let manufacturers collaborate with buyers, forming more accurate forecasts.
  • Reduced costs: Manufacturers can reduce excess inventory by ordering fewer parts before demand is verified.
  • More science, less art: Historical data can be analyzed to improve trend forecasts and to learn from mistakes.

Disadvantages

  • Sudden change: No software can predict massive or unexpected changes in the marketplace.
  • Data pollution: Data may be contaminated by special events.
  • Hoarding: Software is susceptible to companies that order excess inventory to ensure supplies.
  • Buyers' tastes: Software cannot predict market response.
  • Common sense: Software cannot replace experience and insight.

Who Could Benefit

  • Electronics/Technology: Events of the past year have shown the industry's vulnerability.
  • General Manufacturing: Knowing what to build and in what quantity is the ultimate goal of all manufacturers.
  • Retail: The ability to react to changes in the marketplace separates winners from losers.
  • Automotive: Better demand response means lower inventory costs.

Major Players

  • i2 Technologies Inc.:
    The company's supply chain management system is the industry leader.
  • SAP AG:
    The mySAP supply chain management platform has a built-in advanced planning module.
  • International Business Systems AB:
    This Swedish company's supply chain software is strong among companies with international operations.
  • Manugistics Group Inc.:
    Recently added a pricing and revenue optimization engine to its supply chain management platform.
  • J.D. Edwards & Co.:
    Offers an advanced supply chain planning module that works with its core supply chain management offering or third-party offerings.
  • Manhattan Associates Inc.:
    Company's software line offers modules for planning through to fulfillment.

Resources

www.stanford.edu/group/scforum
Stanford Global Supply Chain Management Forum research and discussion portal

www.cpfr.org
Collaborative Planning, Forecasting and Replenishment industry group, developing cross-industry standards

www.ascet.com
Source of white papers on improving supply chain efficiency published by Montgomery Research Inc.

www.ibf.org
Institute of Business Forecasting, organization dedicated to discussing business forecasting

www.supply-chain.org
Supply-Chain Council Inc., a grassroots organization dedicated to supply chain issues

www.imra.org
International Mass Retail Association, providing industry research and guidance