A Demand for Steady Supply

Yossi Sheffi
FT.com / Financial Times
August 22, 2005

Scores of children were disappointed last Christmas as demand for Robosapiens, the must-have toy of 2004, outstripped supply. Many of their parents had their own moment of disappointment, too, as the fashionable iPod mini flew off the shelves faster than it could be replaced. Given the sophistication of modern-day supply chains,such stories are surprisingly common. Yet the central task of the supply chain manager has always been a formidable one: to provide parts and products in the right quantity at the right place and time at the lowest possible cost. Too little, and customers will walk away empty-handed; too much, and discounting and disposal costs mount.

The difficulty is that products are designed, made and ordered before demand is known. And increasingly complex markets make the future even murkier. But managers can reduce forecasting risks by making supply chains more responsive to demand variations.

Creating a demand-responsive supply chain is not easy, but it is possible. These six steps provide a framework.

Range forecasting

Instead of aiming for a single demand forecast, which is invariably wrong, progressive companies have turned to forecasting a range of potential outcomes.

Managers estimate the likely range of future demand, and use the estimates to guide supply contracting terms and contingency plans. More importantly, the practice "socialises" the organisation to expect uncertain outcomes.

Ford, for example, develops a range forecast for product sales as part of its capital asset planning process. The forecasts are updated (monthly, quarterly or yearly, depending on the application), typically narrowing the range as the forecast period draws near.

To cover the range of its forecast, Griffin Manufacturing of Bedford, Massachusetts, uses a multi-sourcing approach. Its Honduras plant is tasked with manufacturing the certain part of the demand, using its low cost to produce high volumes, while the Massachusetts plant, which is more flexible and closer to the market, responds to short-term demand variations.

Risk Pooling

Large-scale or aggregate forecasts are more accurate than individual forecasts. In the late 1990s Cadillac changed its distribution strategy in Florida, one of its largest markets. Instead of allowing dealers to order the cars they assumed customers wanted, Cadillac sent only demonstration vehicles. When a customer placed an order, the car was delivered overnight from Cadillac's distribution centre.

The change allowed Cadillac to pool demand forecasts from its Florida dealers, rather than respond to individual dealers' forecasts. The aggregated forecast was much more accurate than the individual dealers' forecasts and the result was vastly improved customer service. The distribution centre could ship overnight exactly the vehicle that the customer wanted about 75 per cent of the time.

Another common risk-pooling strategy is to reduce the number of product components, a tactic known as part variability reduction. When a company uses common components in lots of its products, it can pool its forecasts for the components' demand. Intel's systems group simplified its usage of parts in this way, reducing 20,000 different part numbers to 500.

Product variability reduction works in a similar way. It is hard to forecast demand for products when there is a huge range of product configurations. For example, the 2000 Mercedes E Class car came in at a staggering 3,900bn possible varieties. This is far beyond what the company can actually stock. Honda, by contrast, offered just 529 combinations of its popular Accord model in 2000 by offering "packages" of options rather than a smorgasbord of potential add-ons. The smaller number of packages allows for better risk-pooling, lower variability and thus better forecasts.

Risk-pooling product variety has its downside, however: it forces customers to select from a smaller number of product variants. The challenge is to make sure they remain satisfied with fewer options.

Shorten horizons

It is easier to predict what will happen tomorrow than what will happen next month. Reducing the time between conceiving, producing and selling a product greatly improves the ability to forecast demand.

To shorten the lead time involved, many companies carry out rapid prototyping to speed up product development, rapid tooling to speed up the ramp-up process and rapid manufacturing to speed the production process. Many of the tasks involved can be developed in parallel and by involving different groups in the company at each step.

Some companies, such as Lucent Technologies, have created a single supply-chain organisation that spans the relevant company functions, including engineering and sales. Others use physical proximity. Luen Thai, a clothing manufacturer, is developing a textile "supply chain city" in the southern Chinese city of Dongguan, allowing customers - such as Liz Claiborne, the fashion label - to shorten the design process and reduce the time from concept to retail store from between 20 and 50 weeks to 12 weeks.

Companies have also successfully combined the ideas of risk-pooling and shortened forecast horizons in a strategy known as postponement or late differentiation. Hewlett-Packard manufactures its popular Deskjet and Deskwriter printers in its Vancouver and Singapore plants, and distributes them in the US, Europe and Asia.

Selling printers in Europe means following each country's requirements for printer configurations: different packaging, power plugs and manuals. In the past, HP forecast demand for each European country and then built the appropriate number of printers for each country. But forecast errors caused frequent surpluses and shortages.

To increase availability without leaving itself with unsold stock, HP switched to pan-European forecasting. It began shipping generic printers to its European distribution centre in Holland, where they were configured for each country once local demand was known.

Test batches

Fashion retailers such as Nine West, the women's shoe store chain, face a common forecasting challenge: each time a new style is introduced demand is unknown.

To improve its forecasts Nine West developed a new process. The first 1,000 pairs of shoes of a new style are flown to five representative US stores where their sales are monitored closely for a few days. That information is then used as an indicator to forecast the sales of the entire line.

The retailer boosts production if sales are above expectations and scales it down if they are disappointing. When a new shoe "bombs" during the tests, Nine West halts production and sends the shoes already made directly to outlet stores and discounters, saving on the costs of transportation to and from its own stores.

Companies can exploit the knowledge of their suppliers and customers to make better forecasts. Wal-Mart is a classic example. In the 1980s the US retailer pioneered a real-time online application that provided its suppliers with up-to-the-minute sales data, helping them and Wal-Mart itself become more efficient.

Good forecasting often involves more than data sharing, however. It requires co-operation between partners to identify discrepancies and amend forecasts. A formal method for doing this is the collaborative planning forecasting and replenishment (CPFR) process, developed by an industry consortium of retailers and packaged consumer goods manufacturers.

In August 2000, Superdrug, the UK retailer, was suffering from mismatches between supply and demand in its health and beauty products business, particularly during sales promotions and new product launches. In August 2000 it launched a CPFR pilot with Johnson & Johnson, resulting in a 13 per cent reduction of Superdrug's inventory levels, an in-store availability improvement of 1.6 per cent, and a better relationship with Johnson & Johnson.

Risk sharing

Sharing the risk of an erroneous forecast with supply chain partners can limit the damage for all parties. Risk sharing can be embedded in supply contracts in many ways, including revenue sharing and options-based contracts.

One of the most common, however, is the buy-back agreement. In the book industry, for example, publishers buy back unsold books from retailers, thus sharing the risk of holding too much inventory and making retailers less conservative in ordering books. Price supports, popular among electronics manufacturers, are a similar mechanism. When the price ofa product falls, due tothe introduction of newmodels, price supports provide for reimbursements from the manufacturer to the retailer.

Such arrangements are often advantageous to both parties because having higher inventory allows the seller to respond more easily to high demand. At the same time, the retailer gets financial support from the manufacturer if the product is not selling and must be discounted below cost. The manufacturer, meanwhile, sells more upfront with a better chance of higher sales. Even if it must bear some of the risk of low sales, its expected profits are higher.

The techniques given above can help companies design a more agile supply chain. Those that master them will create a robust business that is resistant to the inevitable errors of forecasting.

The writer is professor of engineering systems at MIT. He is the author of The Resilient Enterprise, MIT Press, September 2005 (sheffi.mit.edu/resilient-enterprise).