Service parts logistics: a benchmark analysis.

1. Introduction

This paper summarizes and analyzes the findings of a benchmark study of after-sales service logistics systems. Such systems support the provision of service parts, maintenance and repair services to product end users. The study was commissioned by the International

Business Machines Corporation in the fall of 1992 as part of its research partnership with the Center for Manufacturing and Logistics Research at the Wharton School of the University of Pennsylvania. (This Center has been renamed the Fishman-Davidson Center for Service and Operations Management. All inquiries about this research should be directed to the authors at this Center.) The focus of the study was on the computer industry. The study group also included companies from other industries that have particular expertise in the provision of service logistics support. A total of 14 companies participated in the study by providing company data and by sharing their experiences with the project team as well as with each other.

The purpose of the study was to survey current industrial practices and trends in service logistics operations. Specifically, it compared the parts distribution network structure and the control systems utilized by each company in these networks, i.e., inventory stocking policies, information systems, communication systems, and transport modes. The study also examined service measures and standards, as well as cost and revenue performance metrics. The objective was to develop a methodology that could be used to assess the effectiveness of this critical business function within the corporate environment of each company.

The project team and the managers who participated in the study concluded that the project constituted a first step in what should be an ongoing effort. The survey instrument, which was pre-tested and modified throughout the study, provided a means for collecting information that is essential for performance assessment. The analysis methodology explored various approaches for making meaningful, cross-sectional comparisons by defining normalized financial and service-related metrics. Definition of appropriate normalizing factors emerged as an important research question. The specific empirical observations generated here, although preliminary in nature, provide a fascinating picture of the state of the art as it is practiced by some of the leading companies. Our observations of emerging strategies and procedures should be of immediate interest to corporate managers and researchers. Finally, the study illustrates and, to some extent, quantifies the contribution of the after-sales service support function to firm competitiveness in high-technology industries.

The paper is organized as follows. The remainder of this section reviews related benchmarking studies and describes the methodology used. Section 2 summarizes our findings concerning the service logistics environments observed in the sample. Section 3 focuses on performance metrics and their evaluation. Section 4 deals with industry trends and notes some specific improvement opportunities. The final section notes some promising areas for future research. The results presented here consist of highlights of the benchmarking study. A full report of all the data [1] is available.

1.1. Related benchmarking studies

Logistics benchmarking is a widely accepted tool, used by more than 65% of Fortune 1000 companies to improve productivity and quality [2]. There are two basic approaches that have been used to benchmark logistics in general, and after-sales service support in particular. The first approach was made famous by Xerox Corporation's study of L.L. Bean's distribution system [3,4]. This study benchmarked specific activities within the logistics function, such as warehousing, transportation and fleet management. It then compared process efficiency measures, such as lines picked per labor day in a warehouse or the number of pallets placed on a forklift. Such process-focused benchmarking studies are important for improving the efficiency of specific operational steps. Many companies are actively involved in such benchmarking and have dedicated personnel who share information with other companies. The second approach benchmarks a sample of companies with respect to various general competitive aspects of after-sales service, such as profitability, overall costs, customer engineer productivity, and overall customer satisfaction. A number of these more general studies were sponsored by the Association for Service Management International and were carried out by consulting firms [5-8].

The benchmark study reported here extends these studies by focusing on the service parts distribution system as a process. As a result, the structure of this system as well as the control policies and technology in place are studied from a materials management perspective. Data on logistics related costs, inventory investment levels, and customer service were combined with extensive participant feedback.

1.2. Benchmarking methodology

The first step in benchmarking is to select a sample of participants. Given the interests of the study's corporate sponsor (IBM), we picked a variety of computer hardware manufacturers with lines of businesses that include mainframes, mid-range computers, workstations, personal computers, printers, peripherals, and storage devices. We also included an earthmoving equipment company, because it had a reputation for excellent after-sales service. Both industries, moreover, share traits such as the need to maintain technologically complex, expensive equipment, whose down-time can lead to very high customer/owner costs.

The questionnaire was the main data collection instrument of the study, and hence the design of the questionnaire was a very critical task. We designed a comprehensive questionnaire to collect data on operational, financial, and customer service performance of each company's after-sales service logistics function for the year 1992. The instrument also contained questions concerning control policies, system design, and the competitive environment.

The questions were based on a conceptual model of a service logistics system that was derived from managerial input as well as from our own research on normative models of multi-echelon, service parts logistics management systems (see [9-14]). The underlying model framework for the service delivery process is general enough that it can incorporate differences in distribution structure and control policy observed in the companies. However, the model is rich enough that specific distribution network and performance parameters can be defined explicitly. The generic model consists of a multi-echelon service delivery structure in which parts inventory can be deployed at stocking points in many levels This inventory can be allocated to lower echelons that directly satisfy a customer demand, and upper echelons that replenish the lower echelons and also act as an emergency backup. The service delivery process is initiated by a customer call, which may be followed by a visit from the customer engineer (repair person) to the customer site (see Fig. 1). Such calls are typically generated by machine failures, maintenance, and/or engineering changes. The call may require a part to fix the machine. If the part is not available in the field service inventory, and the part is critical, then an emergency shipment is made from an upper inventory echelon. The failed part either can be discarded or can be repaired and put back into the system. Efforts were made in designing the questionnaire to use terminology that is prevalent in the industry. The questionnaire was first tested at the Field Service Group of the IBM Corporation. A group of logistics managers for each of the major product groups at IBM filled out preliminary versions of the questionnaire and suggested improvements to remove ambiguities and to ensure that all relevant factors were included. The revised questionnaire then was sent to the participating companies. Personal interviews were conducted with many of the companies to understand the specifics of their businesses and to clarify their responses to the questions. A follow-up conference was then held to discuss the data and the results of our initial analysis. Representatives of 12 companies attended the meeting. All participants presented an overview of their service parts operations and were given a chance to clarify and/or modify their responses to the questions. As a result of the meeting, follow-up questions were distributed and collected to complement the first questionnaire so that better data quality could be achieved.

[Figure 1 ILLUSTRATION OMITTED]

The benchmark measures introduced here are based on a number of normalized ratios for specific financial and service performance variables. The comparative results based on these measures are especially sensitive to the definition of the particular normalization factor used for each performance variable. The following three normalizing factors were selected to neutralize the effect of company size, technological mix, and organizational structure: (i) the cost of parts used (or consumed), (ii) the level of service related revenues, and (iii) the level of product sales revenues. These normalizing factors acted as surrogates for the level of output of the service logistics function (which should be the appropriate normalizing factor). They were also based on data that are readily available and understood, and their use ensured the confidentiality of the data provided by the participating companies. (All results are presented in an anonymous manner, to preserve respondent confidentiality. All respondents were informed of the location of their own responses on each display.) It is important to note that different normalization factors may be more or less relevant for particular company environments and for different performance metrics. Indeed, the issue of how to normalize cross-sectional performance data remains an important research issue that is being pursued in our follow-up research.

Fig. 2 illustrates the dynamics of the methodology utilized in this study. As noted above, central to our approach is a conceptual model of the service logistics process. Inputs to the multi-step procedure for designing and executing the benchmark include assessment of: (1) current company practice, (2) the service environment of the company, and (3) the company's competitive environment. The outputs of the benchmark include performance measures, indications of best practice, assessment of industry trends, and, finally, an understanding of the causal relationships between performance, policy, environment, and structure.

[Figure 2 ILLUSTRATION OMITTED]

2. Logistics operations in a challenging environment

After-sales service in the computer, communications, and other high-technology industries is facing escalating pressure to improve both the level of service delivered to the customer (end-user of the firm's products) and the productivity in providing this service. Technology advancement and competition in finished product markets have led to a rapid decrease in product price and to the enhancement of product performance. Customer expectations for product reliability have increased. As a result, the provision of superior after-sales service, at a competitive price, has become an important qualifier for competitive survival. At the same time, business environmental changes, such as the rapid introduction of new products and reduction of product life cycles, have increased product variety and thereby reduced the installed base of specific models. These changes have made demand for service parts lower and less predictable. In addition, component miniaturization and the escalating cost of diagnosing product failure have led to increased parts modularity, which, in turn, increases parts cost. As a consequence of all of these factors, the management of the after-sales service function has become more difficult, more expensive and more strategically important.

The following statistics and observations illustrate some of the challenges currently being faced by service delivery organizations (see Table 1 for more background information on the sample's competitive environment).

Table 1. Sample overview

                        Number of                  Sample
                        companies     Range        average

Number of                  12      2500-300 000    85 984
part numbers
Installed machine          10       630-7 million  1.05 million
base (units)
Number of                  10      0-13 000        2 685
customer engineers
Number of                   8      5160-3 million  636 226
service calls (yearly)

* The installed base of products that must be served is large and geographically dispersed. This base averaged 1.05 million machines in the sample.

* There is enormous variety in the number of service parts which must be maintained. On average, 86 000 part numbers are stocked by each company.

* The cost of parts is increasing owing to increasing complexity and modularity. The sample average part cost is $270, with some companies reporting parts that cost hundreds of thousands of dollars.

* Product life cycles are short, which is reflected by the high rate of part obsolescence. On average, 18% of the total operating costs are due to obsolescence and scrap.

* There is a preponderance of slow moving parts. This is due to reduced product sales volumes (per model), which is caused by increased variety and customization. Design improvements have also increased part reliability, which acts to increase the mean time between failures. These trends also contribute to reducing the predictability of parts demands. A part generally remains on the shelf, in the inventory, for more than a year, i.e. inventory turnover is less than one.

In our participant conference it was noted that the emergence of open systems and multi-vendor environments has added to the challenge faced by service logistics management within many companies. More and more customers are demanding that their service providers be able to service products made by different manufacturers. This practice significantly increases the parts variety faced by the support organization. Moreover, this trend further increases competition in the industry, as manufacturers compete with each other and with third-party companies for the after-sales service of their products. (See [15] for an analytical model of this competitive strategy problem.) As noted previously, the cost structure of this function also suggests that the provision of after-sales service can be highly profitable.

2.1. Parts distribution network structure

Companies employ multilevel distribution facility networks to ensure the prompt delivery of parts to a diverse, geographically dispersed customer base. As noted above, many companies are re-engineering their distribution networks to reflect changes to their competitive environments. As a consequence, the sample contained a range of distribution network structures characterized by the number of echelons, the distribution of stocking/service locations over these echelons, and the nature of the supply/demand linkages between the locations.

We observed, in particular, that the three-echelon structure is most prevalent. It is followed in popularity by the two-echelon structure. In most of the three-echelon structures, the middle echelon is dedicated to making emergency shipments only. Therefore, for replenishment purposes, most of the companies use only two echelons. Only two companies use a four-echelon structure. The additional echelon consists of regional distribution centers that act to replenish the lower echelons as well as to satisfy emergency demands. The number of locations in the lowest echelon (local branch offices) for these two companies is much larger (thousands of stocking points) than that observed at other companies operating fewer echelons. It is interesting to observe that, on average, the central warehouses carry only 63% of the part numbers and 57.3% of the inventory value. In contrast, 16% of the part numbers and 36% of the inventory value are held in the lowest echelons. Clearly, fast-moving items are dispersed closer to customers.

All companies struggle with the decision of parts deployment across their logistic networks. (See Agrawal, et al. [16] for a normative model of this decision problem that was motivated by the observations of this study.) In most companies, customer engineers or product dealers decide which parts to stock at field locations (which are closest to the end users). This decentralized decisionmaking process can lead to overstocking of parts at the field locations due to over-ordering or delays in returning unused (surplus) parts. To resolve such inefficiencies, some companies use incentive schemes to induce their customer engineers or dealers to stock the right parts at proper inventory levels. Also, as part of their re-engineering effort, some companies are consolidating field inventory into hubs that can deliver parts directly to customer locations without going through local stocking points. Although this form of centralized control may provide the same parts availability to the customers, it does generate resistance from customer engineers and dealers, who see the change as a loss of direct control over the spares stock.

Finally, we observed that there has been an increase in the use of outsourcing for both warehousing and transportation operations. Although handled physically by a third-party distributor, the service parts are still owned, and often are controlled, by the product manufacturer. Outsourcing of logistics services allows the company to reduce its payroll and its warehousing fixed costs. It also allows them to focus on parts management and customer service. The third-party logistics companies, moreover, often have an advantage over individual companies, owing to the presence of economies of scale and scope.

Replenishment leadtimes have a direct effect on inventory investment. External replenishment leadtimes (i.e., sourced from suppliers outside the distribution network), which vary from part to part, are generally long (see Table 2). These long leadtimes force companies to maintain high safety stocks and to increase their inventory investment. However, internal replenishment leadtimes and emergency leadtimes are much shorter. Quick internal response significantly enhances customer service. In our sample, about 70% of the companies satisfy 95% of their customers' requirements for service parts within 24 hours. Two of the companies provide service to 90% of customer demands within 2 hours. The measure of service used here is the time between placing an order and shipping a part from the stocking point. It does not measure the additional time lag associated with the actual receipt of the part. The relationship between leadtimes and customer service is discussed further in Section 3.1.

Table 2. Range of observed leadtimes

           Regular replenishment       Emergency

External   10 days to 7 months         1-30 days
Internal   2-14 days                   2-24 hours

2.2. Support technology

Communication and transportation technologies can reduce leadtimes and enable the network consolidation and downsizing efforts we noted above. Figs 3 and 4 illustrates the observed usage of communication and transportation technologies.

[Figures 3-4 ILLUSTRATION OMITTED]

Electronic data interchange (EDI) is used widely by the sample companies to support the filling of both stock replenishment orders and customer requirements. This usage has resulted in improved customer service and reduced replenishment leadtimes. Many companies now require their suppliers to use EDI. A number of companies stated that they have been more successful in implementing EDI internally than externally (with suppliers). Consistent with observations from other industries, we noted that barriers to integrating suppliers through EDI arise primarily, as a result of software compatibility problems. Moreover, many smaller suppliers do not have EDI capabilities. These factors contribute to the high external replenishment leadtimes noted above.

A variety of transportation modes are in use in the sample. Trucks currently are the predominant mode for making replenishment shipments. There is, however, increasing use of air shipment for both regular and emergency shipments. Many companies have taken advantage of the competitive air shipment market to deliver emergency orders, nationwide, from a single central warehouse on an overnight basis. Some have started using express couriers, such as Sonicair, for emergency service. Companies using these couriers are able to provide parts to most locations in the United States within 2-4 hours from 4 or 5 emergency parts centers. The higher transportation costs of this mode are reported to have been more than offset by reductions in inventory holding and handling costs for high-cost low-demand parts.

Our survey indicated that basic, understandable inventory management techniques are used widely. Almost all of the companies use the ABC method for parts classification (11 out of 12), min-max stock level controls and some type of forecasting method (all companies). EOQ is also used extensively (8 out of 12). Although most companies use customized inventory management software, about half of them also use DRP or MRP systems for service parts management.

Computer manufacturers have also installed diagnostic tools and software backup capabilities that are triggered by machine failures. These systems can reduce machine downtime and eliminate the need for a customer engineer visit to determine which part should be ordered. Some companies are designing their products for high reliability through the use of redundant and fault-tolerant systems. Finally, we observed that some companies are designing procedures that enable users to service their machines on their own, with telephone assistance and parts supplied by the manufacturer.

3. Performance evaluation and comparisons

Service quality has a direct impact on a firm's market success. In the questionnaire, companies were asked to describe the service performance measures they currently use, as well as the target levels and actual delivery levels for each measure.

Customer satisfaction, of course, is the ultimate measure of service performance. Most of the participants report that they use questionnaires to measure the level of customer satisfaction on a periodic basis. Responses to these surveys also provide feedback on which service dimensions are important for customer satisfaction, and where improvements are required. Clearly, such surveys are critical for helping companies improve their service system's strategic value. Because the survey questionnaires were custom-designed by each individual company, statistics derived from responses would be difficult to use in cross-company comparisons.

3.1. Parts availability and response time

Timely service is important to all of the respondents. This performance variable is more critical to those companies whose products are used on-line to provide a critical function, e.g., central information-processing systems on mainframes, on-line testing equipment in the semiconductor industry and key equipment in large-scale construction projects or in manufacturing operations.

Fill rates are one of the most commonly used measures for parts availability. They are usually measured as the percentage of requests filled immediately from inventory on the shelf. For parts inventory systems with a multi-echelon distribution network in place, a request may be filled at a local stocking point or via an emergency shipment from a remote central warehouse. From a customer satisfaction point of view, the time it takes to fill the request for a part is critical. Therefore, the analysis distinguishes between `fill rate within 2 hours' and `fill rate within 24 hours'. The former measures the fill rate at a local stocking point, and the latter includes fills by emergency shipments from central locations.

Many companies in the sample set their service target by specifying the percentage of customer demands that are met within 24 hours (1 day). This response time was used as a standard for comparison purposes. The data shown in Fig. 5 indicate the distribution of responses. Because timely fill of critical parts is especially relevant for customer satisfaction, the statistics reported here pertain to higher priority parts only. (Parts criticality is, in fact, a complex issue. Most companies equate criticality with the impact of a part's failure on the operation of the product. Subjective methods are used extensively to classify the degree of criticality of individual parts.)

[Figure 5 ILLUSTRATION OMITTED]

The data show that 70% of the respondents satisfy 95% of their customers' requirements for service parts within 24 hours. This illustrates that a high level of service is being provided by the participating companies. We note that the data provided to us consider a request to be filled when an item is removed from inventory and shipped to the customer within the 24 hour time limit. Thus, the additional time lag associated with transporting the part to the customer location is ignored (see Fig. 1). Many companies did not have such data available. For those that did, we observed the difference between the percentage shipped versus received by the customer (within 24 hours) to be within about 2-3%.

For computer systems that support critical business functions, one-day service is often not good enough. Service providers of such products are required to guarantee service within hours of product failure. Therefore, we also collected data on the parts fill rate within 2 hours. This measure also reflects the fraction of demand that is filled without resorting to an emergency shipment from a remote central facility, although it may include the demand filled by emergency shipments through a local courier from a local emergency stocking point. The data indicate that two companies satisfy 90-95% of their demand and another two satisfy 70-90% of their demand within the 2-hour window. It should be noted that some companies also offer service guarantees between the 2-hour and 24-hour standards (i.e., 8 or 12 hours).

3.2. Financial performance measures

After-sales service organizations face increasing pressure to improve financial performance. Many of these organizations are evaluated as profit centers, and hence their return on assets and their inventory turnover level are compared with values for these measures in the manufacturing and finished product distribution functions within the company. All of the participants are especially interested in how financial performance of service operations and service parts inventory management should be measured. They are also keenly interested in how their performances compare with those reported by other companies. Accordingly, this study collected data on each participant's service revenues, operating costs and inventory investment. The data are analyzed by using measures that are normalized to account for the different environment factors noted above.

After-sales service revenues are an important part of the participating companies' total revenues. Such revenues, on average, are equal to 30% of product sales. The average service revenue per customer engineer is $215 000. The majority of these revenues (63%) come from maintenance contracts. Other sources of service revenues include parts sales (5%), carry-in-repairs (2%), warranties (3%), and time and material contracts and internal (within the company) service (27%).

Depending upon their competitive environments, companies will adopt different strategies to maximize profit throughout the user life cycle of the product. For example, some companies charge a high price for the purchase of their products and make little or no profit on after-sales service. Others choose to charge a low price for their products and a high price for after-sales service. The latter companies plan to capture a larger share of this profit by providing superior service during the post-sales, ownership phase. Warranty and service contract terms are additional decision variables that affect the value of the product and company revenues. Hence, the overall pricing strategy of a firm can have a significant impact on revenues derived from after-sales service operations. It is therefore important, to view service function revenues in conjunction with product sales revenues on a full ownership life cycle basis (see [15]).

In this project, service delivery related costs are defined to be the total operating costs associated with directly providing after-sales services. These costs include the opportunity cost of inventory investment as well as logistics (transportation, handling, etc.) costs, overhead and other operating costs. We specifically exclude parts purchase and repair costs. Fig. 6 illustrates the breakdown of the average operating cost structure.

[Figure 6 ILLUSTRATION OMITTED]

The opportunity cost of capital invested in inventory is bounded below by the firm's cost of capital. This cost is usually not reported as a part of the service function's operating costs, although such interest expense is recorded at the corporate level. This cost is included here because it is directly influenced by service logistics management decisions. A common value of 10% was used for the holding cost interest rate and was applied to the reported value of inventory from each company. Inventory value is determined by the purchase cost, to the logistics function, and does not include depreciation adjustments (see the discussion in Section 3.2 below on inventory investment). At this rate, the average holding cost is equal to 28.6% of the total operating costs.

Logistics costs include the cost of transportation (regular and emergency), warehousing, obsolescence, scrap, and shrinkage. They are controllable and are directly related to efficient management of service parts inventory systems. Such logistics costs account for 40.2% of the total operating costs. It is interesting to note that a significant portion of these costs is due to obsolescence, scrap, and shrinkage (17.7% of the total operating costs). The high level of obsolescence costs reflects the impact of reduced product life cycles (which leads to machine replacements in the field) as well as engineering improvements to existing service parts.

Other operating costs include general, administrative, personnel and miscellaneous costs. These costs averaged 31% of total operating costs.

New service parts are acquired either from internal vendors (i.e., the company's manufacturing plants) or from external vendors. On average, companies acquired 22.1% of service parts (as a fraction of total purchase cost) from external suppliers, and 49.7% from internal vendors. A significant portion of the parts (27.2%) is circulated through repair facilities, which may be either internal or external. Because internal transfer prices are often significantly different from the market price of parts, the purchase cost of parts was not added to the operating costs. Similarly, the cost of repair was not included.

The participants' inventory investments (book value) in service parts are significant. On average, service parts inventories equal 8.75% of the value of product sales. It is not a simple matter, however, to compare inventory investment values across companies for a variety of reasons. First, even in the same lines of business, companies often use different accounting procedures (such as LIFO, FIFO, weighted average costing, or market price) to value their inventories. Moreover, some companies depreciate their inventories, whereas others do not. Consequently, we must distinguish between the gross value (based on purchase price) and the net (depreciated) value of the inventory investment. Many companies have data on inventory investment valued by more than one method, and these different values are used for different managerial purposes. To be consistent, we have used the gross value of inventory investment for comparison purposes.

3.3. Turnover and performance measure normalizations

The values of the inventory investment and operating costs are affected by the various environmental factors. Therefore, we present inventory investment and operating costs in the form of various normalized ratios through which comparisons are more meaningful.

Annual inventory turnover is the commonly used measure of efficiency of inventory systems management. This measure indicates the physical speed at which service parts pass through the system. Its reciprocal is the average time parts stay in stock before being used. For example, an annual turnover of 2 indicates that, on average, a part stays in inventory for 1/2 year (six months).

In our sample, inventory turnover is consistently low across the respondent companies (the sample average is 0.87 and the highest observed value is about 2). This low turnover rate for service parts reflects the slow-moving nature of service parts consumption. Their use is based on either the failure of the product in the field or on the consumption of a 'usage' part (e.g., bulldozer blades). We would like to emphasize that low turnover rates for service parts inventory systems are not surprising. Indeed, as noted earlier, they should be significantly lower than the turnovers observed in finished product distribution and manufacturing systems.

It is necessary to use the same accounting method to calculate the numerator and denominator in the turnover calculation. Otherwise, the ratio will not capture the physical turnover of parts. We used annual gross purchase cost of parts leaving the system and the gross value of inventory investment. It is interesting to note that a number of companies use turnover as a performance measure, but apparently are not consistent in their use of accounting methods for the numerator and the denominator. As a consequence, the value of inventory turnover reported to us by these companies was significantly different from that which we computed on the basis of reported inventory and usage data.

It was also observed that for those companies that deal primarily with repairable parts, inventory turnover may have a different meaning, because the 'cost' of a repaired part is ambiguous, i.e., it can be either the cost to repair the part or its original (purchase) cost. Many companies, for convenience, use parts repair costs in the numerator, but use inventory book value (which reflects the purchase cost) in the denominator. `Turnover rates' computed in this fashion no longer represent actual physical movement of goods. This report corrects inconsistencies in usage and inventory data, and thus the turnover results presented here can be used for cross-sectional comparison purposes.

Whereas inventory turnover is a good efficiency measure for service parts inventory systems, its value by itself does not reflect the reasons for holding inventory. All else being equal, a higher demand rate should lead to a higher turnover rate. Fig. 7, which plots inventory turnover against the annual purchase cost for a typical part number (i.e. average cost per item times the average demand in units, per part number), seems to support this hypothesis. The scatter of points in the figure suggests that economies of scale exist and that there is a positive correlation between turnover and scale. It is important to note, however, that the number of observations is small, and hence this conclusion requires further research. (In a recent study, Cohen et al. [17] measured inventory turnover versus sales in the industrial paper and plastic industry. The observed pattern was non-monotonic, i.e., first decreasing and then increasing.) Furthermore, a company must keep a higher parts inventory level to provide better service to its customers. As a consequence, there is a tradeoff to be considered between inventory investment (as measured by turnover) and customer service levels. Other normalized variables can provide insights into the nature of the parts management function.

[Figure 7 ILLUSTRATION OMITTED]

The size of the installed product base that needs to be serviced is a possible choice for a normalization factor. Because our sample includes both very large systems (e.g., mainframe computers) and very small ones (e.g., PCs), we decided that the number of systems in use is not an appropriate normalizing factor. Clearly, the cost of systems in use is relevant. Therefore, a better normalization factor could be based on the current value of the installed machine base. We emphasize `current' here because the prices of computer systems drop very quickly. Moreover, their value in the field should be depreciated and thus will depend on age and usage patterns. Unfortunately, although many of the participating companies have data on the number of machines in their installed base, it is very difficult, if not impossible, for them to report on the current value of this base.

For this report, service revenues and annual product sales (in dollar value) were used as normalization factors for different performance variables. Each of these factors is readily observable and is correlated to the level of output of the after-sales service support function. It is important to note, however, that the normalized measures that result from applying these factors only partly account for the strategic business reasons that affect a company's decision to make a service parts inventory investment.

To capture the benefit of holding inventory, service revenues are divided by the inventory investment. The resulting metric is referred to as the 'service turnover'. This metric is especially appropriate for comparison between service operations that are run as profit centers. The average value of service turnover for our sample is four times per year. This indicates that the companies, on average, generate about $4 of service revenues for each dollar invested in parts inventory.

An argument can be made that the annual sales volume of finished products is also a good surrogate for the value of the installed base of machines. Its use would lead to another turnover type of measure that is equal to the ratio of inventory investment to product sales. Our choice of product sales as the normalizing factor is based on the notion that a firm's sales in the most recent year reflect the current value of the machines as a consequence of rapid depreciation and compressed product life cycles. This measure is also consistent with the fact that investment in service parts inventory enhances service and thus is linked to the firm's finished product sales volume.

In a similar fashion, the total operating costs can be compared across the sample after appropriate normalizations. Service revenue and product sales are both used for this purpose. In our sample, the average of (total cost)/(service revenue) is 39% and the average of (total cost)/(product sales) is 3.32%.

3.4. Cost-service trade-offs and the efficient frontier

In general, the delivery of improved after-sales service requires investment in more resources and the expenditure of more costs. Companies typically set target levels for the level of after-sales service that they hope to attain. Such target levels are based upon a company's business strategy and on its competitive situation. Therefore, cost or inventory investment measures cannot be compared in isolation.

At the very least, a two-dimensional picture of the service-cost or service-inventory investment tradeoff must be considered. A plot of the various normalized measures of each company's cost or inventory investment (as defined above) against the level of service achieved illustrates this tradeoff. The data points that lie along the bottom-right side of the scatter of points define the efficient frontier. Firms on the frontier are the most efficient in that their performances are not dominated by those of any other companies, i.e., they provide maximum service for a given level of cost, or, alternatively, they provide a given level of service at a minimal cost.

The 24-hour parts fill rate is used as the service measure (the horizontal axis) in these graphs. A normalization of total operating costs or inventory investments is used for the vertical axis. Figs 8 and 9 illustrate the results. It is important to note that all of the data are self-reported by the participants, and hence one must use these data carefully in ranking individual company performance. The general shape of the observations is highly consistent, however, with that predicted by inventory theory.

[Figures 8-9 ILLUSTRATION OMITTED]

If a company lies on the efficient frontier of the cost-service graphs, then it certainly is performing at a high level. Otherwise, there may be room for improvement. However, as noted above, the apparent 'inefficiency' of the interior points may be due to differences in the competitive environments or in the organizational structure of the relevant firms. For example, operating as a profit center rather than as a cost center can affect the incentives of a firm to be on the frontier. Similarly, a firm's pricing strategy for both its products and after-sales services can affect this outcome.

Fig. 9 illustrates the tradeoff between inventory investment, normalized by product sales (i.e., 1/service = turnover), and the 24-hour service measure. Because inventory investment is only a part of the resources employed in supporting after-sales service, it should be noted that a company on the efficient frontier on these graphs may not be cost efficient. For example, inventory investment can be reduced by using a faster but more costly transportation mode. It is interesting to note, however, that a great deal of consistency was observed with regard to the placement of each company on both the cost- and inventory-related service tradeoff graphs. Clearly, some companies have developed methodologies and procedures that lead to more efficient parts logistics operations.

4. Management trends and research opportunities

This benchmark study has uncovered a variety of issues associated with the measurement of performance and the management of operations. It has also identified some leading-edge practices that suggest that there are significant opportunities for improvement. Many of these practices are consistent with policy recommendations, for inventory management, which stem from our research experience in multi-echelon systems. This section highlights those opportunities that have significant potential to improve after-sales service logistics performance and concludes with a review of future research directions.

Properly defined performance measures, and incentives consistent with these measures, are essential to support process improvement in the after-sales service function. They are also critical for the definition and adoption of cost effective logistics strategies that are supportive of the firm's overall competitive business strategy. A number of specific managerial recommendations emerged from our analysis.

For example, end-user Customer satisfaction surveys should be used to obtain direct feedback on service performance. Similarly, companies should use service metrics which are directly related to system reliability and `uptime', i.e., job completion rates and response times. (Parts availability is the instrument by which job completion occurs, i.e., high parts fill is a necessary condition for high job completion. See Cohen and Lee [18] and Cohen et al. [10] for a discussion of this measure and formulae that explicitly relate them.) Inventory investment and turnover have emerged as critical internal metrics, and hence they should be computed in a manner that accurately reflects the value of the parts and speed at which they move through the logistics system. Finally, obsolesence costs should be made visible and computed in a manner that is consistent with the inventory valuation procedure.

The participating companies used a diverse set of distribution network structures. Many of these structures are a legacy of past decisions. To adjust to today's environment, companies have begun the process of rationalizing their network structures by reducing the number of echelons and the number of locations at each echelon. This will make the service parts system leaner and more efficient. Savings from such changes include reduction of fixed costs for the facilities that are eliminated.

For highly reliable parts, economies of scale can also arise as a result of risk pooling (variance reduction) achieved by reducing the number of stocking locations. This risk pooling effect is especially beneficial for service parts, because short product life cycles and parts obsolescence make holding stock expensive and risky. It is intuitive that parts with higher costs and lower demand rates should be stocked at central, higher echelon locations. The survey data indicated that the lower echelons accounted for 16% of the parts numbers and 36% of total inventory investment. This suggests that companies may be able to reduce their inventory investment by carefully redesigning stocking policies. Our research in this area has led to the development of analytical models to support such stocking decisions (see [16,19]).

The survey indicated that some companies have not invested in advanced inventory management methodologies. Those companies that did make such investments emerged as benchmarks in the cost-service tradeoff discussed in Section 3.3. Technologies, such as MRP and DRP, which are being used by some, may not be appropriate because they were not designed specifically for the management of service parts. We believe that the following modelling issues are particularly relevant to service parts management and thus should be of interest to inventory researchers: (1) parts classification analysis beyond traditional ABC methods, (2) risk pooling over multiple products, markets, and locations, (3) one for one replenishment policies that impose a `pull' philosophy on stock replenishment, (4) parts stock positioning analysis, (5) repairable item analysis, (6) budget or capacity constraints that restrict the number of parts to be positioned at forward stocking locations (i.e., the `tool-kit' problem), (7) pricing policy analysis, and (8) supply chain coordination across multiple channels and with customers and suppliers.

To the best of our knowledge, this study is the first industrial survey of service parts logistics conducted as a collaborative effort by leading practitioners and inventory management researchers. With its emphasis on processes, the study provides an overview of current industrial practice as well as emerging trends. It has defined relevant performance measures and measured the achieved values of such measures (i.e., to define `best practice' performance). At a policy level, the study indicates how service delivery structure and control systems are changing to respond to changes in the after-sales service environment. The principal managerial contribution, we believe, is to provide actual data on metrics and practices that can serve to stimulate corporate improvement processes.

From a research perspective, this study has identified a number of challenging modeling issues for multi-echelon inventory systems. It has also raised a number of significant questions concerning the causal links among performance, practice and environment. More cross-sectional data must be collected before such links can be analyzed effectively. The benchmark process and associated data analysis methods documented in this paper will, we hope, stimulate further research on defining methods and procedures needed to support empirical analyses of operations management systems.

Acknowledgements

We thank the following companies who participated in this study: Amdahl Corporation, AT&T Network Systems, Caterpillar, Inc., Control Data, Digital Equipment Corporation, Hewlett-Packard Company, IBM Corporation, Panasonic Factory Automation, QMS, Inc., Silicon Graphics, Inc., Tandem Computers, Inc., Teradyne, Inc., Unisys Corporation and United Telephone-Northwest. We also thank the following people from IBM Corporation: Thomas J. Aquino, Chris Crane, J. Blair Fishburn, Susan Otway, John B. Reuter and James Seitz. The contributions of Michael Giffin and Gary Lindeman of IBM, who were deeply involved in the design and execution of the study, are especially noted. We further thank Shuren Zhang for his efforts as a research assistant.

References

[1] Cohen, M.A., Zheng, Y.-S. and Agrawal, V. (1994) Service Parts Logistics Benchmark Study, Center for Manufacturing and Logistics Research, The Wharton School, University of Pennsylvania.

[2] Foster, T.A. (1992) Logistics benchmarking: searching for the best. Distribution, (March), 31-36.

[3] Tucker, F.G., Zivan, S.M. and Camp, R.C. (1987) How to measure yourself against the best. Harvard Business Review, no. 1 (January-February) 8-10.

[4] Camp, R.C. (1994) Benchmarking, in The Logistics Handbook, Robeson, J.F. and Copacino, W.C. (eds.), Free Press, New York, pp. 303-324.

[5] Coopers and Lybrand (1991) Service Operation Strategy of the 90s, Association for Services Management, Fort Myers, FL.

[6] Coopers and Lybrand (1994) Benchmarking Impacting the Bottom Line, Association for Services Management, Fort Myers, FL.

[7] Andersen Consulting (1988) Linking Service Strategy to the Bottom Line, Association for Services Management, Fort Myers, FL.

[8] Andersen Consulting (1990) Future Trends: Prospectives on the International High Technology Services Industries in the 1990s, Association for Services Management, Fort Myers, FL.

[9] Cohen, M.A., Kleindorfer, P. and Lee, H. (1986) Optimal stocking policies for low usage items in multi-echelon inventory systems. Naval Research Logistics, 33, 17-38.

[10] Cohen, M.A., Kleindorfer, P. and Lee, H. (1988) Service Constrained (s,S) inventory systems with priority demand classes and lost sales. Management Science, 34(4).

[11] Cohen, M.A., Kleindorfer, P. and Lee, H. (1992) Multi-item service constrained (s,S) policies for spare parts logistics systems. Naval Research Logistics, 39, 561-577.

[12] Ernst, R. and Cohen, M.A. (1993) Dealer inventory management systems, lie Transactions, 25, 36-49.

[13] Chen, F. and Zheng, Y.-S. (1993) One-warehouse multiretailer systems with centralized stock information. Operations Research, 45(2), 275-287.

[14] Chen, F. and Zheng, Y.-S. (1994) Evaluating echelon stock (R,nQ) policies in serial production/inventory systems with stochastic demand. Management Science, 40(10), 1262-1275.

[15] Cohen, M.A. and Whang, S. (1997) Competing in product and service: a product life-cycle model. Management Science, 43(4), 535-545.

[16] Agrawal, V., Cohen, M.A. and Zheng, Y.-S. (1995) Stock positioning in a multi-echelon service parts distribution network. Working Paper, Fishman-Davidson Center for Service and Operations Management, The Wharton School, University of Pennsylvania.

[17] Cohen, M.A., Agrawal, N., Agrawal, V. and Raman, A. (1995) Analysis of distribution strategies in the industrial paper and plastics industry. Operations Research, 43(1), 6-18.

[18] Cohen, M.A. and Lee, H. (1990) Out of touch with customer needs: spare parts and after sales service. Sloan Management Review, 31, 55-66.

[19] Cohen, M.A., Lee, H., Kamesam, P., Kleindorfer, P. and Tekeran, A. (1990) OPTIMIZER: IBM's multi-echelon inventory system for managing service logistics. Interfaces, 20, 65-82.

Biographies

Morris A. Cohen is the Matsushita Professor of Manufacturing and Logistics in the Operations and Information Management Department of The Wharton School of the University of Pennsylvania. He is also Co-Director of the School's Fishman-Davidson Center for Service and Operations Management. He holds a B.A.Sc. in Engineering Sciences from the University of Toronto, as well as an M.S. in Industrial Engineering and Management Science and a Ph.D. in Operations Research from Northwestern University. Dr Cohen has been an Area Editor of the Manufacturing, Operations and Scheduling Section of Operations Research, and Co-Editor of the Special Issue of Operations Research on `New Directions in Operations Management'. He has also served on the Editorial Boards of Journal of Operations Management, Journal of Manufacturing and Operations Management, Naval Research Logistics, and recently joined the Editorial Advisory Boards of the new journals, Supply Chain Management Review and Journal of Manufacturing and Service Operations Management. Dr Cohen's current research interests include: (1) global manufacturing, logistics, and supply chain management, (2) new product development process management, (3) technology management and automation of value adding processes, (4) channel choice and distribution strategy, (5) service logistics system management, (6) strategic benchmark analysis of operations process, and (7) inventory control and production management. He has been a consultant to a wide range of domestic and global companies on projects that include the development of a model-based analysis of global supply chain strategies for a computer manufacturer, an inventory control system for after-sales service parts distribution in the automobile and computer industries, a decision support system for new product development in the packaged goods industry, and an extensive benchmarking analysis of service logistics in high-technology industries.

Yu-Sheng Zheng is Associate Professor of Operations and Information Management at the Wharton School of the University of Pennsylvania. He holds a B.S. in Mathematics from Fudon University (Shanghai, China), an M.S. in Industrial Engineering from Zhejiang University (Hong Zhou, China), as well as an M.A. in Statistics and a Ph.D. in Operations Research/Management Science from Columbia University. He has been an Associate Editor for Management Science and Operations Research since 1991. Dr Zheng has done consulting for both US and international companies, and has advised major consulting firms on various projects. He is an expert on diagnostic analysis, identifying problems in areas ranging from operations strategies to process re-engineering for complex, large-scale manufacturing and logistic systems. His research interests cover a wide range of design and planning issues in production-distribution systems and supply chain coordination.

Vipul Agrawal is Assistant Professor in the Operations Management Area and Statistics and Operations Research Department at the Stern School of Business, New York University. He holds an M.S. degree in Industrial Engineering from Stanford University, and a Ph.D. in Operations Management from the University of Pennsylvania. Dr Agrawal's current research interests include supply chain design, supply contracts, after-sales service logistics, benchmarking, and electronic commerce.

MORRIS A. COHEN, YU-SHENG ZHENG and VIPUL AGRAWAL

Fishman-Davidson Center for Service and Operations Management, The Wharton School, University of Pennsylvania, 3209 Steinberg Hall- Dietrich Hall, Philadelphia, PA 19104.6366, USA

Received July 1995 and accepted September 1996