Steps to Successful Segmentation

Some researchers confuse cluster analysis with segmentation. Cluster analysis is a mathematical tool used to divide consumers into groups that are similar to each other in some way. Segmentation is a marketing strategy that targets marketing initiatives against subsets of consumers in order to achieve an improved position in the market (higher share, penetration, loyalty, usage, etc.). Therefore, cluster analysis is a tool to help marketers achieve segmentation. For segmentation to be successful, one must follow all the steps:

1. Clustering
2. Statistically testing the clusters
3. Profiling the clusters to determine if they satisfy the segmentation objectives

1. Cluster Analysis Cluster analysis is a collection of statistical methods used to assign cases to groups. Cluster analysis methods will always produce a grouping, which may or may not prove useful for classifying objects. If the groupings discriminate between variables not used to do the grouping, and those discriminations are useful, then cluster analysis is useful. For example, if grouping zip code areas into fifteen categories based on age, gender, education, and income discriminates between wine drinking behaviors, it would be very useful information if one were interested in expanding a wine store into new areas.

There are two primary forms of clustering tools: hierarchical and non-hierarchical methods. Hierarchical methods build clusters by combining the two most similar respondents or clusters together in a stepwise approach. Non-hierarchical methods attempt to partition the data by splitting the data into subsets. In addition to these two methods, hybrid approaches may also be followed.

Issues to Resolve Before Choosing a Clustering Procedure Before employing a particular clustering technique, one must consider the following: What data should be used to form the clusters? How should similarity be measured and how should clusters be formed? What method should be used to determine the optimum number of clusters? Answers to these questions will eliminate some clustering techniques from consideration. Other techniques will be eliminated because they introduce so much compromise that actionable segments will not emerge. (All methodologies depend on levels of compromise. Perhaps the source of the greatest amount of compromise is in the number of clusters formed. If too few are formed, they might not be differentiated enough to support a targeting effort; if too many are formed, they might not be large enough to be viable targets.)

2. Methods to Statistically Test Clusters Techniques such as discriminant function analysis, CHAID ( chi-square automatic interaction detector) , and C&RT (classification and regression trees) can be used to statistically test and describe the structure of the clusters. In other words, one can determine how different the clusters are from one another. If the clusters are sufficiently unique, they will be further described and become named segments.

3. Profiling Clusters to Determine if Segmentation Objectives are Satisfied Of all the complex analytic tasks involved in segmentation research, profiling of clusters is by far the easiest. By examining the properties of the cluster it is possible to develop a robust description of the cluster. Simple cross-tabs with a cluster banner can be used to identify cluster differences on demographic, lifestyle, life stage, geographic, behavioral, attitudinal, or other variables used to target. Profiling makes the clusters come alive and become segments.