Having identified the relevant segmentation variables with structural or value chain implications, the next task is to combine them into an overall segmentation of the industry. The task is usually difficult because there are many relevant segmentation variables—in some industries there can be dozens. The challenge is to distill these variables into the most meaningful segments for developing competitive strategy.

The first step in the distillation process is to apply a significance test to each segmentation variable. Only those variables with a truly *significant *impact on the sources of competitive advantage or industry structure should be isolated for strategic analysis. Other less important, though still meaningful, segmentation variables that are identified can be used for fine tuning in marketing or operations management.

The basic tool for translating the remaining variables into a segmentation is the *industry segmentation matrix. *A simple segmentation matrix based on two segmentation variables is shown in Figure 7-3, illustrating the oil field equipment industry in which the size of the buyer oil company and the stage of development of the country in which the buyer is headquartered have been identified as the two segmentation variables.

The first practical problem in constructing a segmentation matrix is choosing the number of categories of each segmentation variable to select. In Figure 7-3 I have chosen three discrete categories of buyer size and two categories of a country’s stage of development. In reality, buyer size is a continuous variable and country development goes through many stages. The way in which each segmentation variable is broken into discrete categories should reflect the categories that capture the most significant structural or value chain differences, balanced against the practical need to limit the number of segments to a manageable number. Deciding on the best discrete categories for strategic purposes almost always requires judgment and is an iterative process.

The cells in Figure 7-3 are the individual segments in the industry. It may well be that some of the cells are presently unoccupied. In addition, if there were no small independent oil companies based in developing countries and not likely ever to be any, this segment would be a null cell. For purposes of illustration, Figure 7-3 shows null cells involving both large and small independents. Segments can often be eliminated from consideration if they are null cells. However, it is important to remember that null cells should be *infeasible *combinations of the segmentation variables and not merely cells in which no firm is currently operating. Feasible cells where no firm is operating represent a potential opportunity and it is important that such segments be highlighted, not eliminated, in segmentation.

Figure 7-3. A Simple Industry Segmentation Matrix for an Oil Field Equipment Industry

Figure 7-3 portrays a case where there are two relevant segmentation variables. In practice, there may be many variables grouped under the four broad categories of product, buyer type, channel, and geography. Looked at closely, most industries are quite heterogeneous. With many significant segmentation variables, the number of segmentation matrices that could be plotted multiplies rapidly. The problem, then, is to convert the segmentation variables into a small number of segmentation matrices that will be most illuminating for the strategy formulation process.

### 1. Relationships Among Segmentation Variables

To move from a number of segmentation variables to the most meaningful segmentation matrices, the first step is to probe the relationships among the segmentation variables. The number of important segmentation variables can be reduced by collapsing segmentation variables together that are correlated, or which effectively measure the same thing. For example, geographic location may be associated with a particular buyer type (e.g., automobile companies are located in the Midwest), or buyer type may be closely related to channel (small roofing contractors are all served through distributors). Constructing a segmentation matrix with correlated segmentation variables will produce a matrix in which many cells are null.

Segmentation variables that are highly correlated can be combined, because one variable is a surrogate for the effect of the other. In less extreme cases, the correlation among segmentation variables is partial, but allows a significant reduction in the number of possible segments because many cells in the matrix are null. It is important to identify all the relationships among the segmentation variables and use this to combine variables together and identify null cells.

It is also important to understand *why *variables are related, because this will often have important ramifications. If one variable is not a good surrogate for another but rather a reflection of current firm behavior or happenstance, combining variables is a mistake. It will obscure unoccupied segments that may represent an unexploited opportunity. For example, if small roofing contractors were served through distributors not for economic reasons but for historical rea- sons, then eliminating direct sale to small contractors as a segment would be a mistake. Telemarketing or remote order entry by salespersons with portable computer terminals might make the segment feasible though it had not been previously.

### 2. Combining Segmentation Matrices

The significant and independent segmentation variables that remain after the process described above represent the potential axes for industry segmentation matrices. Where there are more than two segmentation variables, the industry segmentation matrix will no longer fit on a two-dimensional page. One way of proceeding is to construct a number of different segmentation matrices for each pair of variables. Each of these matrices can then be analyzed for its strategic implications. This approach is not fully satisfactory, however, because meaningful segments may be the result of combining more than two segmentation variables and may be overlooked.

To deal with more than two segmentation variables, it is usually useful to create *combined *segmentation matrices. The process is illustrated in Figure 7-4. In oil field equipment there are at least two other relevant buyer segmentation variables besides buyer type and geographic buyer location: the technological sophistication of the oil company and its ownership. In Figure 7-4, I have plotted the four variables in pairs and then combined the two segmentations together after eliminating null cells.

The process of combining matrices not only reduces the number of segments by eliminating some null cells, but also exposes correlations among variables that may have been missed. In Figure 7-4, I have noted the null cells representing infeasible combinations. Combining matrices is usually best done by combining all segmentation variables within a category first. In Figure 7-4, for example, I have combined all the buyer segmentation variables together.

After combining segmentation variables of the same broad cate-gory, one proceeds to combine variables in different categories. In doing so, it is usually best to create a segmentation matrix in which one axis reflects the combined *product *segmentation variables and the other axis combines all the *buyer-related *variables (buyer type, channel, geography). Where the number of segmentation variables is manageable, it is possible using this procedure to construct one twodimensional industry segmentation matrix. This matrix may be quite large, but has the advantage of displaying the entire industry in a way that facilitates strategic analysis. Figure 7-5 shows such a matrix for oil field equipment, after adding to the segmentation two product segmentation variables—premium versus standard quality products, and products with ratings for deep versus shallow drilling.

Sometimes the number of relevant segmentation variables and resulting segments is so great as to make a single matrix unwieldy. The presence of a very large overall segmentation matrix should prompt the reexamination of the segmentation variables and the discrete categories of each to ensure that the differences are truly significant. Where this is the case, it may be desirable to use two or three segmentation matrices in subsequent analysis to avoid missing important strategic implications.

The industry segmentation matrix should contain potential segments and not just segments that are currently occupied. Potential segments may imply entirely new segmentation variables (e.g., channel is added because there is the possibility that some direct sales may be possible in the future instead of handling all sales through distributors) or new discrete categories of existing variables (e.g., a new performance rating for an alloy).

A segmentation matrix is an analytical tool, not an end in itself. The analyst should start with the longest list of segmentation variables to avoid overlooking possibilities. Only over the course of the analysis are variables combined or eliminated and the working segmentation matrix refined. The whole process usually involves trying a number of different segmentation schemes in which the product and buyer differences that are most important for industry structure are gradually exposed.

A segmentation matrix should be tested by examining the strategies of competitors. If the scope of competitors’ activities is plotted on the matrix, new segments or segmentation variables may be exposed. Conversely, competitors’ activities may draw attention to segments that must inevitably be served together. I will have more to say about this below when interrelationships among segments are discussed. Figure 7-6 summarizes the steps required in industry segmentation.

Source: Porter Michael E. (1998), *Competitive Advantage: Creating and Sustaining Superior Performance*, Free Press; Illustrated edition.