Printer Friendly

Factory focus: segmenting markets from an operations Perspective.


Market segmentation techniques are routinely used by marketers to divide markets into groups of customers showing similar buying behavior. This practice has enabled businesses to implement targeted marketing strategies that yield important benefits such as: the formulation of efficient and effective marketing programs designed and directed to groups of customers with similar buying characteristics, improved product pricing through the development of separate pricing structures, and products designed especially for different market segments.

While it is obvious to marketing managers that a single business can serve multiple market segments, each exhibiting different buying behavior and requiring a different marketing strategy, it is often not apparent to either marketing or operations managers that it is difficult for a single operating unit to develop superior competitive capabilities in serving market segments that require very different operations strategies. For example, recently the manufacturing director at a packaging materials plant faced three different market segments. One group of customers required the frequent delivery of small lots on a JIT basis within a 24- to 36-hour time interval. In another segment, the customers were highly price sensitive, requiring low prices for large orders, while the customers in the last segment placed large orders but required a premium quality level with strict adherence to delivery reliability against pre-set schedules. In order to be profitable in each of these segments, different types of investments in processes and infrastructure are required; i.e., different manufacturing strategies. Yet the manufacturing strategy for this plant focused on the low cost production of high volume products and did not recognize the differences in requirements between the three market segments.

Our field research indicates that operations and marketing managers often fail to recognize that different customer needs can dictate different operations capabilities in a plant. Frequently, operations managers view their task in generic terms covering a single requirement for their business as a whole, such as the achievement of low cost manufacturing or time-based manufacturing in all areas of the business. While such approaches to operations may effectively support parts of their business, they may also offer little competitive advantage in others. Likewise, marketing managers are often unaware of unique operations capabilities that can differentiate the firm's products from their competitors' in particular market segments, thereby missing sales opportunities. The potential loss of business, the magnitude of the investments in operations, and the resulting time frame over which these investments impact the business all argue the critical need for understanding market segment differences with regard to operations capabilities.

This paper addresses the role of market differences as they affect the development of an appropriate operations strategy. In particular, we discuss the problem of segmenting markets with regard to operations capabilities, and the application of cluster analysis to market segmentation, using operations-related variables. We report the results of field research designed to test this approach. The cluster analysis results for two manufacturing firms and one service firm illustrate the importance of market segment information in the formulation of operations strategy.


The concept of "Focused Factories" has been articulated in the manufacturing strategy literature as a means of making the manufacturing function an important competitive weapon by outstanding accomplishment on one or more measures of manufacturing performance (Hayes and Wheelwright (1984), Hill (1989), Skinner (1969, 1974). While this is clearly an important concept for operations managers, what has not been made clear is just how to determine what focus is necessary to best support the markets targeted by a particular business. We believe market segmentation is an effective technique for determining plant focus. Moreover, we believe that Skinner's original argument in developing the focused factory concept supports this view, "that focused manufacturing must be derived from an explicitly defined corporate strategy that has its roots in a corporate marketing plan, and that the choice of focus cannot be made independently by production people" (Skinner (1974)).

The manufacturing strategy framework reported by Hill provides one way of linking marketing and manufacturing perspectives during the formulation of corporate strategy, and of integrating marketing and operations perspectives (Hill (1989)). In particular, Hill's use of order winners, qualifiers, and order-losing sensitive qualifiers to define the key manufacturing tasks in different market segments provides a means of specifying the data required to segment markets for operations purposes. Qualifiers are customer requirements for which a certain level of performance is necessary from all companies competing for the business. Reliable delivery and minimum quality levels are common examples.

Qualifiers to which a customer would be particularly sensitive, such as cleanliness or chemical purity standards, are designated as order-losing sensitive. Despite their importance, meeting qualifier requirements will not, by itself, win the order for a firm. Requirements on which superior performance would differentiate one firm from the others, such as price or delivery speed, are referred to as order winners.

In this study, we use Hill's framework and extend his concept of order winners for a market segment to include additional key operational variables. These order winners and additional variables are then incorporated into a clustering-based methodology designed to help management define the operations task for different segments of the market. An important advantage of the approach is that it permits market segments to be defined which may, in fact, be very different from those defined for the same market using traditional marketing approaches, thereby providing important operations perspectives on the business. Furthermore, viewing a market in operations terms can actually enrich the strategic alternatives considered by marketing.


There are major differences between the use of market segmentation techniques, such as cluster analysis, to define market segments for the purpose of developing marketing strategy, and their use to define the operations task in different market segments. One key difference is in the variables considered by marketing and operations. In marketing, surrogates for buying behavior are frequently used. For consumer products these include demographic variables such as age and income. For industrial products, variables such as company size, SIC code and location are often used. In either case, what marketing strategists are concerned with is identifying important buyer behavior differences between segments so that marketing strategies can be developed which exploit these differences. For operations, the key variables to be studied are those that relate to identifying differences in the operations processes and infrastructure used to support different market segments. The following operations variables are typically of concern:

* Market factors defining buyer behavior. How does a company win orders for its products/services in relevant markets? What operations capabilities are required to compete in the market?

* Time factors. Due to the size and fixed nature of the investments in operations, it is essential for a business to know how the relevant market factors change over time.

From this market view, three critical operations issues need to be identified:

* Process technology factors. What process technologies and skills are needed to make the product or provide the service?

* Volume factors. What order sizes are processed? A knowledge of production/operations volumes is frequently critical in making process choice decisions.

* Infrastructure. What investments need to be made regarding key facets of infrastructure, including systems, procedures, controls, people, and organizational structure?

Another key difference is that, while marketing and operations will necessarily consider data for the same customers and products or services in a given market, the objectives pursued by these two functions differ significantly and can result in very different ways of segmenting the same market. Furthermore, the fact that customers are divided into different groupings for the development of marketing and operations strategies is not important. What is most critical is that marketing and operations managers understand and are able to explain the different views of the market taken by the two functions. Different perspectives on market requirements, and an awareness of the capabilities required by the firm to support these requirements, are important preconditions for the development of coherent strategic plans for a company.


We propose a methodology that uses cluster analysis to categorize a set of products or customer orders that are representative of a business's markets, based on their similarity or dissimilarity across a set of operations variables. These orders and their characteristics are identified through discussion with general management and key marketing and operations executives, using the methods reported by Hill (1989). Data concerning order winners and qualifiers, operations volumes, and processes are essential for this analysis. However, the goal of the clustering-based methodology is to ensure that operations differences in a firm's set of orders are clearly identified; in effect, to "break up" operations into focused product groups.

We use a two-step clustering approach that has been recommended and validated in the literature (Dillon and Goldstein (1984), Pung and Stewart (1983)). In the first step, a hierarchical clustering model is used to generate a dendogram that graphically illustrates the linking patterns between sample orders. Based on the results of the first step, we then use a K-means clustering model to identify the major operations-based segments found in a business's markets.(1)

In our research, we have found that the relative emphasis put on different operations variables will change to reflect the relevant market dimensions faced by a certain business. We report field research results in this paper which indicate the wide range of operations variables that may be considered in a market segmentation analysis. To illustrate this point, Table 1 lists the particular operations variables of importance to the three firms discussed in this paper. Clearly, management may identify many variables that apply in a given situation. It may also consider some variables to be more important than others when forming focused product groups (or market segments as defined by operations variables). As a result, unaided attempts to compare multiple products along a large number of key variables may be difficult, if not impossible. The advantage of the proposed methodology is that it can form product groups in a manner that is consistent with management's views.
Factors Company A Company B Company C
Industry Manufacturing Service Manufacturing
Order-Winners X X X
Considerations X X
Volume Levels X X
Qualifiers X X
Qualifiers X X
Requirements X


In the remaining sections of the paper we report on the application of cluster analysis to data gathered through extensive interviews at three companies.(2) These companies represent three very different industries: electronics, publishing, and packaging. The operations variables studied in these companies for market segmentation purposes include: order winners, a mixture of qualifiers, order-losing sensitive qualifiers, planning horizon considerations, as well as volume and process considerations. Table 1 indicates the key operations variables at each firm, and also shows the increasing complexity of the market segmentation problem as we progress from Company A to C.

In the following sections we briefly discuss each company, paying particular attention to the key variables involved. We describe how these variables were incorporated into the model, and present a methodology-derived solution. This solution is then compared and contrasted with the current situation of the company. Finally, we draw general conclusions concerning the utility of the proposed methodology, and describe the important differences in the operations strategies required to support the various market segments found at each company.


Company A is a manufacturer of printed circuit board assemblies. The company serves several markets, including television, radio, video, appliance, and rending machine products. Five years ago the manufacturing organization was divided into four distinct units using product type and life cycle criteria. These autonomous units included: 1) a pilot line for the development and introduction of new products, 2) a "catch-all" manufacturing unit, 3) a unit designed to handle circuit board requirements for the vending machine market, and 4) a spares unit. In the past, products had remained in the manufacturing units to which they were originally assigned, despite changes in both volume levels and market requirements (as measured by order winners). Recently, management decided to reexamine the product groupings because of declining delivery performance and a shortage of skilled employees for the pilot line.

Key Variables

Management identified, through market analysis, three order winners that characterized demand for their products. In this instance they were: price, delivery speed, and quality.(3) Two sets of order-winner weights were estimated for 28 products considered to be representative of the company's market; one based on current market requirements, and another based on projected market requirements in two years. Progression along the product life cycle was expected to cause the estimated order-winner weights for several products to change considerably in the next two years. This was caused by the fact that Company A supplied current products to original equipment manufacturers (OEMs), as well as spare parts for items that had been discontinued from the OEMs' regular production.

Order volumes varied dramatically within the sample. The projected average weekly production times in two years ranged from 34.8 minutes per week for one spare part item to over 913 hours per week for a new product. The net result is that management had three factors to consider: order winners, shifts in order winners over time, and volume differences between products. Table 2 shows the projected order winners and volume measures for the representative sample of products, and their current manufacturing unit assignments.

Coding and Assigning Weights to the Key Variables

The data for each product consist of four variables projected over the next two years: price, delivery speed, and quality order winners; and production volume. Order-winner weights ranged from zero to 100, with a maximum of 100 points per order. The weights were based on management's judgements and were checked against available data. Because of important volume differences between the OEM and spares business, the projected average weekly production requirements were used to classify products as high or low volume products. If a product was considered to be a high volume item (mean weekly production |greater than or equal to~ 90 hours), then the volume variable was given a value of 200. Otherwise, this variable was set to 0. This rather heavy emphasis on volume is consistent with the process alternatives faced by Company A. For higher volume circuit boards, manufacturing can effectively use equipment with faster cycle times but longer (and more expensive) setups. For the lower volume items (such as spares), a slower, but more flexible, process having shorter setup times may prove better suited to meeting market needs.

When using cluster analysis in this context, it is important to recognize that the weightings are meant to capture differences in the order of magnitude between variables, not exact numerical differences. Viewed in this light, the weightings used for Company A reflect management's feeling that volume is significantly more important for identifying operations-based market segments than are the other three variables.

Statistical Results

Figure 1 shows the dendogram generated by the hierarchical clustering model (Dillon and Goldstein (1984), Pung (1983)). The dendogram indicates that the sample products quickly cluster into three to five clusters. Also, a comparison of the product groupings shown in the dendogram with the current company data in Table 2 reveals that the clusters cleanly break into high and low volume groupings.

Using the dendogram as a guide, the K-means model was run using the four key variables mentioned above as grouping variables (Dillon and Goldstein (1984), Pung (1983)). Three iterations of the K-means model were performed, with the number of clusters set at three, four, and five. A comparison of the solutions indicates that cluster membership was stable across all three solutions; that is, new clusters were formed only by splitting apart larger clusters. In addition, the F-ratios for the key variables indicated that, for the three- and four-cluster solutions, only the delivery speed order winner did not show significant differences (p = .05 level) between product groups. Delivery speed did, however, become significant in the five-cluster solution.

Manufacturing Implications

The three-cluster K-means solution is shown in Table 3. Each of the clusters represents a potential unit with a focused manufacturing task. Segment 2 includes high volume products sold to price sensitive customers, while Segment 1 requires short delivery lead times to customers who buy low volume OEM products and spares. Segment 3 includes premium quality products sold to price sensitive customers who buy low volume products. A comparison of the order winners and production volume data for the current manufacturing units in Table 2 and the recommended manufacturing units in Table 3 (which are based on organizing a manufacturing unit for each of the different market segments) suggests that significant changes in the TABULAR DATA OMITTED organization of manufacturing could lead to a major improvement in manufacturing focus.

The reassignment of products to focused manufacturing units would address the regression in manufacturing locus that has occurred (Hill and Duke-Woolley (1983)). It would also permit management to assign highly skilled operators to Units 1 and 3 where their skills would be most appropriate due to short runs, frequent changeovers, and high quality requirements. The recent increase in the need for highly skilled employees was caused by the major increase in volume experienced by maturing products that were still assigned to the "pilot line." The reorganization of manufacturing units would also mean that the requirement for low cost production in Unit 2 could be supported with lower skilled operators, and major efforts could be placed on reducing process cycle times in order to improve run time productivity. The reorganization of the production units would also enable manufacturing to better serve customers requiring short delivery lead times. Currently, short lead times are often obtained only by incurring extra changeover costs for the high volume, price sensitive work in the "pilot" and "other" production units.


The potential for improved consistency in the manufacturing units is supported by the fact that, for all four key variables in this analysis, the cluster analysis solution has a lower average standard deviation across the units than the company's current manufacturing organization--despite the fact that the cluster analysis solution consists of three manufacturing units instead of four. The results also demonstrate that focusing operations through the use of focused manufacturing units does not necessarily mean a proliferation in the number of units.

These results illustrate why a different segmentation basis is frequently required by operations in order to develop a consistent manufacturing task for individual operating units. They show how segmenting a company's market with regard to key manufacturing variables can help management develop a manufacturing strategy which focuses on the specific mission in each of the firm's market segments.


Company B is a high-speed copying service with four retail locations serving university communities in two states. The services provided by the company range from self-service convenience copiers to more elaborate back-room work, such as litigation and architectural jobs, flyers, and customized textbooks for university courses. The customized textbooks are prepared by instructors, and include materials such as cases, class notes, laboratory manuals, etc. They account for approximately 40% of the annual copying volume at Company B and around 50% of annual sales revenue. The demand for the customized textbooks is highly concentrated in two periods each year, August/September and December/January.

The top executives at Company B decided to take a closer look at this market. Because of the relative size of the customized textbook segment and the growing signs of increasing competition, management was interested in identifying the important customer requirements in this market, and in understanding what operations capabilities were required to meet these customer needs.

Key Variables

Using a representative sample of 34 customized textbook orders, management was able to identify 13 distinct market requirements that they felt explained why customers bought from Company B. Of these 13, six had direct implications for operations: delivery speed, delivery reliability, product quality, price, store sales support, and marketing programs. The market programs variable reflects customer interest in a gift certificate or price discount which a customer could receive by placing orders prior to the peak demand period. These marketing incentives affect operations by shifting demand into non-peak periods, but also increase cost pressures on Company B's operations to pay for the marketing programs.

It is interesting to note that, since the processes required to produce customized textbooks have minimal setup times, management at Company B did not see order volume as being a significant operations variable. This is in contrast to Company A, and reinforces our earlier point that the set of relevant key variables depends on the nature of the business and the characteristics of the process alternatives.

Coding and Assigning Weights to the Key Variables

Table 4 contains the data for each sample order with regard to the six operations-related market requirements. In contrast with Company A, the market requirement variables for Company B include qualifiers (represented by "Q") and order-losing sensitive qualifiers ("QQ"), as well as order winners. Due to the high level of price awareness in the market, management felt price was a qualifier for all 34 sample orders.(4) Management also identified marketing programs as a qualifier or order-losing sensitive qualifier for three of the sample orders.


The scheme used to weight order winners for Company B differs somewhat from the one used in Company A and Company C, to be discussed next. For each order, management used the following values to rate how important each of the potential order winners was to winning the business:

0 - Unimportant 25 - Somewhat Important 50 - Important 75 - Very Important 100 - Vital

For Company B, there was no limit to the number of points that could be assigned to an orders. Nevertheless, three of the orders in Table 4 have no weight given to any of the variables listed. Company B won these orders due to its performance on some factor not directly related to operations, such as advertising.

Since the clustering weights determine the relative importance of a variable, qualifiers and order-losing sensitive qualifiers were given a value of 200 while order winners were scaled from 0 to 100. The logic behind this is straightforward: superior performance on order winners can differentiate a firm from its competitors, but satisfactory performance on qualifiers is essential just to compete in a market segment.

Statistical Results

Figure 2 shows the dendogram generated by the hierarchical clustering model. From the dendogram it appears that the sample orders group relatively quickly into three or four main clusters. Using the dendogram as a guide, we ran the K-means model using the five key variables as grouping variables. Three iterations of the K-means model were performed with the number of clusters set at three, four, and five. A comparison of the solutions shows that new clusters were formed only by splitting apart larger clusters. We can therefore conclude that the clusters identified by the K-means model represent a robust solution. In addition, the F-ratios for the key variables indicate that, for all three solutions, only the store support order winner does not show significant differences (p = .05 level) between order groups.

Operations Implications

The customized textbook market is normally viewed from a marketing perspective at Company B. This includes market segments relating to customer type (professor or administrative staff), end use (academic department or course type and level), or product type (laboratory manuals, collections of research papers, and casebooks). While these segments may be very useful in developing marketing strategy, they do not provide insights into those customer needs supported by operations.

A comparison of the four-cluster K-means solution shown in Table 5 with the original customer order sample shown in Table 4 indicates that, from an operations perspective, the market has three important segments. These include: 1) a delivery speed segment, 2) two groupings of orders that are sensitive to the marketing programs, and 3) a large delivery reliability and quality segment. The customers in the delivery speed segment expect to place their orders during the peak demand period and to have customized textbooks available for students within one week of placing the request. In order to support these requirements, operations is scheduled three shifts per day, seven days per week during the peak demand period. Furthermore, limitations are placed on the scheduling of other university and corporate work during the peak period in order to devote about one third of the normal daily capacity to delivery speed sensitive customized textbook work.

The marketing programs segment (Groups 2 and 4) includes customers who place orders at least one month prior to the peak demand period, and also contains a substantial portion of the firm's new customers. This is an important segment for operations. Sales in this segment permit a substantial increase in the volume of work accepted, and reduce the cost of order processing since the orders can be run prior to the peak demand period. Sales in this segment also release a substantial amount of peak period capacity, thereby allowing Company B to better support the delivery speed segment. However, much more accurate demand forecasts are required for these orders, since under-production places additional demands on capacity during the peak period and over-production produces excessive scrap costs.

Customers in the delivery reliability and quality segment place orders with the company because of the firm's product quality and delivery reliability performance. Supporting these requirements means 1) incurring extra costs to provide proofreadings and quality checks beyond those provided by the firm's competitors, and 2) adding systems for operations planning and scheduling to ensure the reliable delivery of customer orders on the promised dates. Achieving dependable deliveries also means pulling greater emphasis on developing systems and data that result in realistic customer delivery date promises, thereby avoiding an overload on peak period capacity.

Segmenting the customized textbook market from a marketing perspective does not provide a clear vision of the operations capabilities required to be competitive in the customized textbook market. On the other hand, segmenting from an operations perspective reveals important differences in the operations strategy required to win business in each of the segments. Two of the groups also underscore the value of marketing programs in achieving sales revenue growth and improved margin objectives in the business.


Company C manufactures a wide variety of customized molded plastic containers. Currently, management sees the firm's markets as including six main segments defined in terms of end TABULAR DATA OMITTED product use: food products, assembled products, industrial products, fluid containers, cosmetic products, and sundry products. Products within each of these segments require some combination of molding, printing, and assembly processes. Manufacturing at Company C is currently organized according to process (molding, printing, and assembly) rather than by product type. The one exception is the production of fluid products, which has dedicated molding, printing and assembly processes.

Recent decisions to invest in two new molding presses and an additional printer have led to a reevaluation of how manufacturing is organized. Management has already established that manufacturing for cosmetic and food products (items 1-4 and 21-23 in Table 6) be combined since both groups require cleanliness procedures that would be too expensive to replicate. Organization of the remaining product lines, however, remains undecided.

Key Variables

Management identified the important competitive factors for a representative sample of 27 products from each of the firm's six market segments. For each product, the data included current and four year projections of order winners, qualifiers, and order-losing sensitive qualifiers. As Table 6 shows, four years from now, products are expected to differ widely with regard to requirements for price, print quality, and mold design. Production run lengths are also expected to vary considerably in the sample, ranging from a projected 30 standard hours for product 15 to over 10,000 standard hours for product 18 in four years.

There are important differences between Company C's products and those of Companies A and B with regard to process investment. Since both molding and printing represent a substantial investment in equipment and trained personnel, it is important to avoid unnecessary costs with regard to these capabilities. Since not all products require printing, it is important to consider the requirements for this process when forming focused product groupings. Three of the segments (foods, cosmetics, and fluid containers) have products which require printing.


The data for each product considered in the cluster analysis consisted of seven key variables reflecting the order winners and qualifiers in four years: price, delivery speed, delivery reliability, mold quality, print quality, cleanliness, and mold design. In addition, there were two variables that reflect important process requirements: production volume and printing requirements. The order winners were weighted in the same manner as in the Company A analysis, with a maximum of 100 points for each order. Qualifiers and order-losing sensitive qualifiers were assigned values of 200.

The volume variable was used to classify products as either high or low volume products. Products with average run lengths of 1000 standard hours or more were given a value of 200, while lower volume products had this variable set equal to zero. The printing variable was set to 400 if a product required printing; otherwise, it equalled zero. Based on this weighting scheme, the clustering techniques consider the key variables in the following order of descending importance: 1) printing requirements, 2) average order size, qualifiers and order-losing sensitive qualifiers, and 3) order-winners.


Statistical Results

Since food and cosmetics products had already been grouped together due to special cleanliness requirements, the clustering was performed using sample products from the remaining four groups only. Figure 3 shows the dendogram generated by the hierarchical clustering model. The dendogram indicates that the sample products in these four groups do not rapidly form into a relatively small number of clusters, as in the cases of Companies A and B. However, using the dendogram as a guide, we performed the K-means model analysis using the nine key operations variables at Company C as grouping variables. Three iterations of the K-means model were run with the number of clusters set at three through five. A comparison of the solutions indicates that the cluster membership was stable across all three solutions with new clusters formed only by splitting apart larger clusters. Table 7 shows the food/cosmetics products unit and the remaining three manufacturing units generated by the three-cluster K-means model. The table shows a clear separation between products according to printing requirements and production volume levels.

Manufacturing Implications

An examination of Table 7 indicates four market segments with distinct manufacturing requirements:

* Segment 1 requires process conditions that support the cleanliness requirement for food and cosmetic products as well as a printing capability. The key order winners include delivery speed, mold quality, and price; delivery reliability and print quality are important qualifiers.

* Segment 2 includes low volume assembled, industrial, and sundry products not requiring printing. Price is the key order winner, and delivery reliability and mold quality are important qualifiers.

* Segment 3 includes high volume fluid container products requiring printing, with price and delivery speed as the key order winners, and delivery reliability, mold and print quality as important qualifiers.

* Segment 4 includes high volume products that do not require printing. Delivery speed, mold quality, and price are key order winners with delivery reliability as an important qualifier.

The cluster analysis results suggest one basis for developing focused manufacturing units that have limited objectives and are consistent with the market segments served by the company. Organizing a separate manufacturing unit to support Segment 2 would not require an investment in additional printing capacity. The principal task for this unit would be low cost manufacturing in order to provide the margins required to be competitive in the market. In this case the TABULAR DATA OMITTED implementation of a work shift pattern, involving a four crew/seven day workweek, would provide a cost effective approach to manufacturing by maximizing the utilization of equipment. Moreover, an emphasis on reducing the changeover times for orders would also reduce cost and increase manufacturing flexibility for the small volume work in this unit.

The organization of separate production units for Segments 3 and 4 is a debatable issue. In both cases the market requirement for delivery speed means that a four crew/seven day work shift pattern would be inappropriate since a conventional three shift/five day workweek provides the flexibility to increase production volume on short notice by working weekend overtime. Likewise, technical skills to support mold quality are required for both markets. However, an investment in printing is not required to support Segment 4. Furthermore, an emphasis on low cost manufacturing is more important in supporting the fluid container products in Segment 3 than in Segment 4. Given sufficient sales volume in Segments 3 and 4, separate production units may be appropriate.


The results reported in this paper demonstrate the feasibility of using market segmentation analysis to characterize the markets served by a company in terms of the requirements placed on operations. Such an analysis can highlight differences in the operations task required by different segments, enabling managers to debate the effectiveness of various process and infrastructure choices when formulating the operations strategy. Moreover, this approach enables marketing managers to take into account both the investment in operations required to support different marketing initiatives, and to consider how existing operations strengths can provide competitive advantage.

This approach is most useful in companies that serve multiple market segments from a single plant, a situation typical of many businesses. For example, in Companies A and C, the cluster analysis results provided the basis for manufacturing unit recommendations that are far more focused in task than the current manufacturing organizations. In the case of Company B, the results clearly indicate important differences in the operations task for the market segments currently targeted by the company, thus enabling appropriate strategic response.

It is important to note that cluster analysis has traditionally been viewed as a descriptive tool, and indeed, at one level we have used it to uncover various segments already present in the customer sets of three different companies. However, to the extent that cluster analysis can help management create new operating units which are better focused than the old ones, the methodology described here has important implications for strategic management decisions.

Several additional benefits are derived from using cluster analysis to segment markets from an operations perspective. First, the results indicate that this form of analysis can easily handle a large number of conceptually different operations variables, including order winners, qualifiers, process technology requirements, and production volumes. Second, the application of cluster analysis on a regular basis can provide important information to management indicating focus regression in operating units, i.e., a gradual unfocusing of operating units over time (Hill and Duke-Woolley (1983)). By calculating and plotting the standard deviation values for the operations variables in each operating unit (e.g., the orders winners), an increase in these values over time can indicate an unfavorable trend toward focus regression. Finally, the application of cluster analysis for market segmentation in operations can be quite important in situations where the set of relevant variables is very complex, or where inconsistences or differences of management opinion need to be brought into the open.

The concept of focus provides many companies with an approach for developing operations capabilities that directly support their business requirements. Given the dynamic and competitive nature of today's markets, many firms also need approaches that will help them recognize market differences over time. The work described in this paper offers a way of gaining this level of insight as a prerequisite to the implementation of the focused factory approach.


1. See the appendix for a detailed description of the clustering methodology used here.

2. The analysis was run on a Macintosh SE using the statistical package SYSTAT, version 3.2.

3. Delivery speed is defined as a "company's ability to deliver more quickly than its competitors" or the ability "to meet the delivery date, when only some or even none of the competition can do so" (Hill (1989, pp. 40-42)). Quality is defined as conformance to specifications.

4. Since price is a qualifier for all orders, it is omitted from the resulting clustering analysis.

5. As Dillon and Goldstein note, "|a~ variety of suggestions have been offered as to how to form the K starting points used as initial estimates of cluster centers." The interested reader is referred to Dillon and Goldstein for a treatment of this issue.


Dillon, William R., and Matthew Goldstein. Multivariate Analysis: Methods and Applications. New York: John Wiley & Sons, 1984.

Hayes, Robert, and Steven Wheelwright. Restoring Our Competitive Edge. New York: John Wiley & Sons, 1984.

Hill, Terence, Manufacturing Strategy: Text and Cases. Homewood, IL: Irwin, 1989.

Hill, Terence, and R.M.J. Duke-Woolley. "Progression or Regression in Facilities Focus." Strategic Management Journal, vol. 4, 1983, 109-121.

Pung, Girish, and David Stewart "Cluster Analysis in Marketing Research: Review and Suggestions for Application." Journal of Marketing Research, vol. 20, May 1983, 134-148.

Skinner, Wickham. "Manufacturing: Missing Link in Corporate Strategy." Harvard Business Review, May-June 1969, 136-145.

Skinner, Wickham. "The Focused Factory." Harvard Business Review, May-June 1974, 113-121.



The following example illustrates the steps in using cluster analysis to break an operation up into focused product groups. Table A1 presents measures of three order winners for six representative products from a printed circuit board manufacturer's markets. The associated weights are based on management's view of the competitive requirements for each product. The data in this example include only six of 28 products in the full sample, and only three of the order winners actually measured. As a result, the total order winner weights for each product may be less than 100. These simplifications were made in order to illustrate the application of cluster analysis to the data.
Product Price Deliver Speed Quality
 1 80 20 0
 2 80 0 20
 3 60 40 0
 4 50 25 25
 5 30 0 30
 6 40 0 30

Step 1

The first step is to use an agglomerative hierarchical clustering model to group the products according to their similarity across the key variables. The term "agglomerative" refers to the fact that the model starts out with n observations, and continues linking observations or groups of observations until all observations have been assigned to a single group. "Hierarchical refers to the notion that the allocation of an object to a cluster is irrevocable, i.e., once an object is assigned to a cluster it is never removed and joined with other objects belonging to some other cluster" (Dillon and Goldstein (1984)).

In order to use a hierarchical model, a measure of similarity (i.e., a distance metric) between observations and a linking method must be chosen. The distance metric literally represents the p-space distance between two observations, where p represents the number of key variables. The linking method then uses these distances and a joining rule to determine which observations or clusters will be linked together, and in what order. The methodology proposed in this study uses an average Euclidian distance metric, defined below:

|Mathematical Expression Omitted~

where |X.sub.ik~ represents the value of the kth key variable for observation i. Using average Euclidian distances, the distance between Product 1 and Product 2 is:

|Mathematical Expression Omitted~

Table A2 below shows the Euclidian distances between the six products along the three order winners.
 1 2 3 4 5
 2 16.33
Product 3 20.00 28.28
 4 22.73 22.73 17.80
 5 35.59 29.44 33.67 18.71
 6 31.09 23.80 31.09 15.81 5.77

We will use an average linkage method to form the clusters. This method works by computing the average distance between all possible pairs of objects in different clusters, and joining the observations/clusters with the smallest average distance. Since no observations have been linked yet (i.e., all clusters are of size one), the shortest distance between any two observations determines the first linkage. In this case, Products 5 and 6 are linked together to form the first cluster, referred to below as Cluster A.

Distances are then recomputed, except now the distance between any product and Cluster A is equal to the average of the product's distances to both 5 and 6 For example, the distance between Product 2 anti Cluster A is:

|d.sub.2.A~ = (|d.sub.25~ + |d.sub.26~)/2 = (29.44 - 23.80)/2 = 26.62

and the new distance table is shown in Table A3.
 1 2 3 4
 2 16.33
Product/ 3 20.00 28.28
Cluster 4 22.73 22.73 17.80
 A 33.34 26.62 32.38 17.26

Following the logic of the average linkage method, the next linkage would be Products 1 and 2, with a distance value of 16.33. We will refer to this new cluster as Cluster B. Under the hierarchical approach, this process continues until all the observations have been eventually linked into one cluster. Multiple cluster solutions are found by looking at the intermediate linkage results prior to this final step. The order of linkages can be graphically represented by a dendogram, or tree diagram. The dendogram shown in Figure A1 indicates the linkage pattern for out example.

The dendogram has several advantages. First, it gives a general idea of the

similarity between products and clusters. For instance, Products 5 and 6 appear to be more similar than Products 3 and 4, while Cluster A (products 5 and 6) is more similar to Product 4 than Cluster B (Products 1 and 2) is to Product 3. A more important advantage of the dendogram is that it gives an initial idea of the number of "natural" clusters inherent in the data. Note in Figure A1 that since distance is represented horizontally, drawing a vertical line through the dendogram can show how many clusters have been formed within a certain amount of distance. Specifically, the number of horizontal lines intersected by the vertical line indicates the number of clusters present at that level. If strong natural clusters are present in the data, clusters should form fairly quickly and only a few clusters should be present at relatively small distances.

In many cases, however, there is not a clear set of clusters, and more than one solution may be feasible. Variance-based measures have been proposed (Pung and Stewart (1983)), but evaluating solutions still remains a largely subjective task. In our example, Lines Y and Z represent two possible solutions. Line Y represents a two-cluster solution with Products 1, 2, and 3 in one cluster, and Products 4, 5, and 6 in the other. Conversely, line Z represents a three-cluster solution formed by breaking off Product 3 into its own cluster. Average order winner values for the clusters in each solution are shown in Table A4.


The two-cluster solution has one product group dominated by price considerations (average Price = 73), and another group with moderate price and quality requirements. The three-cluster solution has two similar groups, plus an additional cluster containing one product that competes on both price and delivery speed.

Step 2

The second step uses an iterative-based clustering technique to generate a final solution starting from an initial solution found in Step 1. While conceptually similar to hierarchical clustering models, iterative-based methods start with a set number of clusters and use heuristics to reassign observations into clusters. So while a five-cluster hierarchical solution is found by "breaking apart" a four-cluster solution, an iterative-based model may rearrange all the objects in a given size cluster solution.

The specific type of iterative-based method used in this study is known as a K-means approach. The K-means approach works as follows. The algorithm chooses K observations as the initial estimates of cluster centers, where K equals the expected number of clusters.(5) Observations are assigned to minimize the Euclidian distance (or "error") between the observation and the mean vector for a cluster. As a result, the mean vector will change as observations are assigned to clusters, and reassignment of some observations will take place in an effort to minimize total error among the observations. Observations can be moved "from one cluster to another until no transfer of an individual observation results in a reduction in the (total error)" (Dillon and Goldstein (1984)).

For our purposes, there are three basic advantages to using both a hierarchical and iterative-based clustering model:

1. The K-means approach requires the user to have an idea of the number of actual clusters. As the dendogram above shows, this information can be readily supplied by a hierarchical model.

2. Since the clusters identified by the K-means approach can change considerably for different K values, varying K can give us an idea of the robustness of the product groups. Specifically, the more stable cluster membership is over different values of K, the more homogeneous and distinct (i.e, focused) the clusters are likely to be.

3. Using both types of models can serve as a reliability check for the clustering solution(s).

Tables A5a and A5b show the K-means results for two and three clusters, respectively. The tables indicate that cluster membership was stable across both solutions. That is, the three cluster solution is equal to the two cluster solution except for Product 2, which was broken out of a larger group to form the third cluster.

Focusing attention on Tables A5a and A5b, we can begin to see the usefulness of the proposed methodology. For the two cluster solution, computed F-statistics for the key variables show that price was the only order winner to show significant differences between clusters at the .10 level. Put another way, the results imply that price, more so than delivery speed or quality, is a potentially important factor for management to consider when separating products into two focused manufacturing units. Average order-winner values for each of the clusters indicate that products in cluster 1 tend to be price sensitive, while cluster 2 products have only moderate price and quality requirements. The three cluster solution in Table A5b offers an interesting contrast. For the three cluster K-means solution, delivery speed appears to be the most significant differentiator between clusters, with price being a close second. This occurs since all products with delivery speed requirements are contained in just one cluster, while price and quality requirements are reflected in all three clusters.



It should be noted that the K-means results are somewhat different than the two and three cluster solutions generated by the hierarchical model. In particular, the two cluster solutions differ with regard to cluster membership of Product 4, while the three cluster solutions have several differences in cluster members. This is not completely unexpected, given the differences in the algorithms used by the two clustering methods.
COPYRIGHT 1991 American Production and Inventory Control Society Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1991 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:Special Issue on Linking Strategy Formulation in Marketing and Operations: Empirical Research
Author:Berry, William L.; Bozarth, Cecil C.; Hill, Terence J.; Klompmaker, Jay E.
Publication:Journal of Operations Management
Date:Jul 1, 1991
Previous Article:Time and product variety competition in the book distribution industry.
Next Article:Management of technology in service firms.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters