April 1996 € Volume 23 € Number 2

Shape Up, Ship Out

How a team of OR analysts redefined manufacturing strategy for the IBM PC Company in Europe, and saved the computer giant $40 million per year in distribution costs.

By Gerald Feigin, Chae An, Daniel Connors and Ian Crawford

In early 1993, the European arm of the IBM PC Company faced a number of challenges that posed serious threats to its ability to remain a major player in the competitive European personal computer market. Against a backdrop of continuing world-wide recession, IBM PC Company Europe faced unrelenting pressure from increasingly agile and aggressive competitors which were eroding IBM's market share. Pressure came in the form of frequent price cuts, rapid customer order response times, and a steady arrival of new products and features. Poor forecasting only added to the problem, leading to critical shortages of popular products and excess supplies of others.

At the same time, IBM's new CEO, Louis Gerstner, reacting to record corporate losses, made clear the importance of reducing operational costs and inventory throughout the corporation. Under particular scrutiny was the largest of IBM's personal computer manufacturing plants, a sprawling 1.3 million square foot factory located in Greenock, 30 miles west of Glasgow in the heart of Scotland's famed Silicon Glen. The plant, comprised of manufacturing, warehouse and staging areas, had produced approximately 1.2 million personal computers the previous year and was responsible for supplying the bulk of IBM's personal computers to Europe, the Middle East and Africa.

Recognizing the urgency of the situation, as well as the difficulties in finding the right balance between cutting costs and maintaining customer responsiveness, the IBM PC Company management enlisted a team of operations research analysts and management science experts from IBM's T.J. Watson Research Center in Yorktown Heights, N.Y., to help identify, analyze and recommend the most cost-effective changes for IBM Europe's PC manufacturing and distribution operations. Working closely with the PC Company executives responsible for manufacturing and distribution strategy in Europe, the team (which included three of the four co-authors of this article) developed a supply chain simulation model designed to quantitatively assess the impact of various operational strategies on cost and customer service levels.

With this model, the team was able to compare different manufacturing execution strategies, examine the effect of different planning and forecasting methods, and identify lower cost distribution policies. Ultimately, the analysis led to significant changes in both manufacturing and distribution, including the adoption of a build to order (BTO) manufacturing strategy, a direct-ship distribution process that by-passed costly country distribution centers, and the rejection of a cost-inefficient idea that had been gaining currency among IBM executives -- the introduction of a new late-customization assembly plant on the European continent.

The Analysis
The first step in the analysis was to develop a detailed understanding of manufacturing and distribution as it was currently practiced in the PC company. We needed to understand all of the relevant processes in place before developing a model. These included planning processes, manufacturing execution processes, and finished goods distribution processes. Based on this first stage of analysis, the team identified key areas it believed to have a major impact on operational costs and customer responsiveness. In the process of understanding the existing business processes, we generated many ideas on how to improve them. Some of these ideas were original; others were borrowed from operations research and management science literature.

We next constructed a detailed simulation model (see accompanying story on the Supply Chain Simulation Model) of the relevant business processes so that we would have a platform for evaluating proposed changes and directly assessing their effect on costs and service levels. We first validated the simulation model against the existing business policies and then explored alternative policies. We viewed the simulation model as a vehicle for convincing both ourselves and PC company management that proposed changes in operational policies were appropriate.

Demand and Supply Planning Activities
The PC Company's planning activities begin with the demand planning process, which attempts to generate an unbiased forecast of expected shipments of each product in each time period (typically monthly or quarterly). The forecasts are based on historical sales information, product announcements, market and economic conditions, competition, etc. The demand forecast is then translated into two types of detailed plans via a material requirements planning (MRP) system: the production requirements at Greenock, called the "base plan," and the component requirements on the parts suppliers through a bill of materials explosion.

A large number of parameters govern this translation including lead times, frozen zones (period in which previous orders cannot be modified), minimum thresholds for order quantities and changes in orders, and safety stock levels. These parameters are set either by specific part numbers or according to part groupings based on annual usage. One of the supply planning activities is to assess the capability of the suppliers to produce a feasible production and procurement plan. A typical planning cycle used to take at least one full month.

We recognized that supply planning was critical in determining Greenock's parts inventory and its responsiveness to customer orders. We decided to analyze several changes to the supply planning process that we believed might improve customer service levels and reduce inventory holding costs. The first approach focused on modifying the way in which safety stock levels were set. In Greenock's MRP system, uncertainty in demands were dealt with by adjusting the product demands to be above their average values by a certain fixed percentage (a common practice among production planners).

We believed that significant reduction in inventory would be possible by implementing an algorithm to plan the requirements for parts that achieves a specified service level for products while attempting to minimize the safety-stocks of parts required. The algorithm, developed by operations research analysts at IBM, exploits commonality of parts as well as differences in price and lead time reliability between suppliers. While benefits of component commonality in reducing overall inventory are well known, this algorithm, referred to here as Flex Planning [1], was among the first to be developed that addressed industrial size problems.

The second modification of the supply planning process focused on the issue of constrained supply. When not enough parts are available from suppliers to build the forecasted demand, an allocation of the constrained parts to products must be performed. This allocation process is usually performed in an ad hoc manner because MRP systems assume the supply of parts to be unconstrained. In our analysis, we wanted to assess the value of performing this allocation in an optimal way, by solving an optimization problem. To do this, we utilized the Production Resource Manager (PRM) [2], also developed by OR practitioners at IBM, which can perform this optimization based on many different business objectives, including product priority, fair allocation and revenue maximization.

The results of our simulation studies indicated that the Flex Planning method for safety stock planning required less inventory compared with the existing MRP-based method to achieve the same service level (see Figure 1). As for the use of the Production Resource Manager to handle constrained parts, our analysis indicated that it is marginally useful when there are normal, minor shortages; but when there are significant parts shortages, its use becomes critical. Today, Greenock is one of several IBM sites deploying PRM-based applications for planning processes.

Figure 1

Manufacturing Execution
Greenock operated in a Build-To-Plan (BTP) mode in which it built products according to the base plan (or forecast), pushing finished products into warehouses and country distribution centers. There were several perceived advantages of the BTP strategy. For products having large and stable demands, manufacturing production schedules for these products could be implemented so as to use manufacturing resources efficiently. Provided that the right product was on hand in the finished goods warehouse, IBM could quickly respond to a customer order by drawing upon that finished goods inventory. The major drawback of the BTP strategy was that it provided no flexibility for meeting actual customer orders. If the forecasts and subsequent base plans were not accurate (as was often the case), manufacturing would end up building the wrong products, thus tying up assets in finished goods that were not wanted by customers.

We believed that IBM would benefit from a more responsive manufacturing strategy. We developed two alternative approaches which we compared to BTP via simulation: Build- To-Order (BTO) and Late-Customization (LC). We defined BTO to mean that a product is built from its subassemblies only after an order for that product has been received. Under the LC strategy, products are built from shells. A shell is a partially built PC that consists of such sub-assemblies as a power supply, a frame, planars or motherboards, and minimal memory and disk storage. There may be several types of shells depending on the final configurations of the finished products. The shells are built according to the base plan (which in this case specifies the demand for shells) and are stocked at the manufacturing warehouse. Final assembly for products using the LC strategy is triggered by firm orders.

The major advantage of the BTO strategy is that assets are not tied up in unwanted finished goods; PCs are built only to real orders. The BTO strategy can exploit the commonality of components that may exist across the different products. Rather than committing a common critical component to a product whose demand has been forecasted, as in the BTP case, the BTO strategy does not commit critical components until the order is firm. The major drawback to BTO is its effect on manufacturing efficiency and serviceability. With no finished goods inventory to draw upon under the BTO strategy, a product must be built from its subassemblies and tested after the order has been received.

For customer orders requiring rapid delivery times, it may not be possible to build, test and ship a product to a customer within the requested time. Manufacturing efficiency may also be affected by a BTO strategy if large setup times are incurred switching from the building of one product to a different product. Finally, order volatility, also known as order skew, may adversely affect the performance and utilization of a BTO manufacturing line. It is common for order volumes to surge at the end of month, the end of the quarter and the end of the year.

The LC approach is a compromise between BTO and BTP. Its advantages include a possible smaller final assembly and test time compared to the BTO approach, and a more efficient use of manufacturing capacity since the shells are built to a plan. It may also provide a less costly means to meet service requirements in outlying regions. On the other hand, the LC approach has its disadvantages. Shells are built to a base plan so assets may be tied up in certain types of shells which are not needed. Also, the multiple-stage build process may incur additional handling and storage costs, and may expose the products to more handling damages.

Our analysis revealed that under the then current business environment, the BTO strategy achieves the same level of service with significantly less inventory in the supply chain than both the LC and BTP strategies (see Figure 1). Based on this analysis, we strongly recommended a shift to a BTO strategy -- a recommendation that was accepted and implemented. To handle seasonal demands that exceed Greenock's capacity, IBM has contracted with a local vendor to perform some assembly operations when needed. After the initial implementation of the BTO strategy, Greenock adopted a hybrid BTO/BTP process in which standard off-the-shelf products are built to plan. This modification allowed Greenock to be more responsive to high-order skew.

It is worth noting that a BTO strategy is not necessarily the best for all manufacturing environments. Also, the choice of manufacturing strategy -- BTP, LC or BTO -- does not lessen the importance of forecasting and part planning. In an industry in which the procurement lead times for critical components are typically several weeks or months, and product life cycles on the order of several months, good procurement planning is crucial for the success of manufacturing performance. The BTO, LC and BTP strategies all fail if the right components are not available when needed.

Finished Goods Distribution
Prior to 1993, IBM's PC distribution network in Europe consisted of IBM-managed country distribution centers and transshipment points. The distribution centers served as warehouses and staging areas for configuring orders. Customers, consisting mostly of independently owned dealerships and retail outlets, placed orders which were filled by the distribution center in the country from which the order originated. A typical order consisted of some number of system units, together with monitors, country-specific keyboards, cables and documentation in specified languages. Other peripheral devices such as printers and storage backup devices might be included in orders as well.

Once an order was configured, it would be shipped from the country distribution center to the customer. The distribution centers would receive replenishments from IBM Greenock. These shipments usually proceeded by truck from Greenock to a nearby port, then by ship to one of 13 transshipment points located throughout Europe, and then by truck to the country distribution centers.

Several concerns were raised about the existing distribution process. First, service levels -- measured as the fraction of on-time order shipments -- were low. Distribution centers often did not have the right finished goods on hand to fill customer orders and thus had to wait for shipments from Greenock. Shipment delays, especially at transshipment points, were not uncommon. Second, the costs of the distribution network were high. Freight rates that were negotiated were not necessarily the best, the operational costs of the transshipment points and country distribution centers were high, and significant inventory was being held by the distribution centers. Third, IBM believed that increasing competition would necessitate meeting service requirements that could not be met at a reasonable cost given the existing network.

As a result, IBM was seriously considering the creation of a final assembly or late-customization plant -- located closer to primary European markets -- which would perform such functions as adding additional cards or hard disks to the system units, loading language specific software, and bundling peripheral devices with the system units. The plant would handle customer orders with stringent service requirements that could not be met at a reasonable cost directly from Greenock. In addition, a reduction in the number of transshipment points and distribution centers in Europe was expected to significantly reduce distribution costs and delivery times. Finally, questions were raised about whether order consolidation could be done more efficiently in Greenock and shipped directly to dealers (or even the end customers), rather than at the distribution centers; and whether the shipping of orders to dealers would be better managed by IBM or by a commercial shipping vendor.

Using the supply chain simulation, we were able to create a detailed model of the distribution process that enabled us to test various distribution strategies and to compare their performance in terms of total distribution costs and the achieved service level. We used the simulation to evaluate two different distribution strategies for the European market. In Strategy 1, customer orders are consolidated in Greenock and shipped directly to dealers via IBM's network of transshipment points. Orders would be shipped in bulk to these transshipment points and then shipped individually to the final customers. In Strategy 2, order consolidation occurs at two locations: in Greenock and at a proposed late-customization plant located on the European continent.

In our analysis of Strategy 1, we considered several cases in which we varied the number of transshipment points in Europe (See Figure 2). A rough cut location optimization analysis followed by a detailed simulation revealed that the least expensive distribution strategy involved the use of three transshipment points located at strategic ports in Europe. When we evaluated cases with as many as 13 transshipment points, we found that the volume of orders shipped to any one site was not sufficient to permit frequent shipments from Greenock. Instead, orders would be held at the shipping dock until a container was full enough to justify a shipment. Thus, significant delays were incurred and inventory holding costs were large.

Figure 2

The costs of operating the distribution network with three transshipment points were significantly smaller due primarily to lower operational costs and inventory holding costs. Also, we observed, somewhat unexpectedly, that service levels improved slightly when we reduced the number of transshipment points because of the increased frequency of shipments to each of the locations. When we further reduced the number of transshipment points to fewer than three, overall costs began to increase mainly because of the large inland freight charges incurred when shipping orders from the transshipment points to the customers.

In our evaluation of Strategy 2, we were able to clearly quantify the tradeoff in service level and costs associated with introducing a late-customization plant in Europe (see Figure 3). In particular, the simulation results indicated that introducing the plant only made sense if more than 25 percent of all orders required fulfillment in under two days. Because the market demand for such stringent service was much smaller than 25 percent, we concluded that the introduction of an additional plant was not warranted.

Figure 3

As a result of our analysis, we recommended that IBM direct ship orders from Greenock to customers, bypassing country distribution centers. The estimated savings in distribution related costs was approximately $40 million per year. By avoiding the country distribution centers and effectively pooling inventory in one location, IBM would improve its customer service levels and, at the same time, decrease finished goods inventory.

Lessons Learned
In addition to the insight provided by our analysis, IBM benefited from our work in another significant way: By providing an objective means to quantitatively assess the impact of different decisions, we helped to defuse a politically sensitive and emotionally clouded decision process. As a result, potentially harmful but politically easy decisions were avoided and good, but less popular, decisions were easier to make. While the analysis we performed and the recommendations we made have been instrumental in making IBM's manufacturing and distribution operations more competitive, the external factors that led to IBM's problems -- stiff competition, short product life cycles and volatile customer demand -- are still very much present in today's market.

As the battle to increase market share -- and the efforts to streamline operations, reduce costs and improve customer service -- continues today and into the future, operations research and management science professionals in IBM will play an important role in introducing new and innovative operational methodologies and assessing the impact of policy changes on overall business performance.

References
1. R. Srinivasan, R. Jayaraman, R. Roundy and S. Tayur, "Procurement of Common Components in a Stochastic Environment," IBM Research Report, RC-18580, 12/1992, 45 pages.

2. B. Dietrich, D. Connors, T. Ervolina, J.P. Fasano, G. Lin, R. Srinivasan, R. Wittrock, R. Jayaraman, "Production and Procurement Planning Under Resource Availability Constraints and Demand Variability," IBM Research Report, RC-19948, 2/1995, 40 pages.

Gerald Feigin, Chae An and Daniel Connors are Research Staff Members at the IBM T.J. Watson Research Center in Yorktown Heights, N.Y. They and their colleagues in the Production, Distribution and Transportation Research Department work on applied operations research problems. Ian Crawford is the Director of Manufacturing and Distribution for the IBM PC Company Europe. He is also the plant manager of the IBM Greenock PC plant. The authors acknowledge the help of Steve Buckley, Ranga Jayaraman, Tony Levas, Nitin Nayak, Raja Petrakian and Ramesh Srinivasan from IBM Research, and Alan Peat and Ian Baillie from IBM Greenock.

For more information about this article, input the number 5 in the appropriate space on the Reader Service Form

E-mail to the Editorial Department of OR/MS Today: orms@lionhrtpub.com

Lionheart Publishing, Inc.
2555 Cumberland Parkway, Suite 299, Atlanta, GA 30339 USA
Phone: 770-431-0867 | Fax: 770-432-6969
E-mail: lpi@lionhrtpub.com