Take a Look at Your Customer Clusters

Marketing researchers have a very useful technique for dividing your customers, or potential customers, into groups that you can then target for special treatment.

Segmentation analysis, often referred to as cluster analysis, is used to identify homogeneous subgroups within a general population. The term represents a broad category of statistical procedures commonly used for partitioning a market into groups of potential customers who have similar needs or characteristics and who are likely to exhibit similar purchasing and/or usage behavior.

Segments are created based on similarities among different types of variables, which can include (but are not limited to) demographic, geographic, psychographic, and behavioral variables. Ideally, data cases are grouped into clusters interdependently so that within-cluster variation is minimized while between-group variation is maximized. In other words, clustered cases are not only similar to each other, but also dissimilar to cases outside of the cluster.

In this article, we will examine the differences between a priori and post hoc segmentation, look more deeply at three types of post hoc segmentation, and examine the end-uses of a segmentation analysis for clients.

A Priori vs. Post Hoc Segmentation

A Priori segmentation is not based on any empirical research specific to the segments being created. Rather, segments are established by using readily available, pre-existing classification such as Standard Industry Classification (SIC) groups, VALS (Values and Life Styles), or even simple demographics. While it is a cheaper option, a priori research can be dangerous because of the unstable nature of many market segments � while a set of segments may have existed strongly in data collected five years ago, they could easily now be obsolete.

Post Hoc segmentation, also known as cluster analysis, is based on empirical research conducted specifically for outlining market segments. While more expensive, post hoc research is more likely to produce accurate, actionable results. However, since the segmentability of a data set is largely unknown until actual data is collected, there is always a risk that a large about of money may be spent on non-actionable research.

Types of Post Hoc Segmentation

There are three popular approaches to post hoc cluster analysis: hierarchical clustering, k-means clustering, and two-step clustering.

Hierarchical clustering is generally used for smaller samples of less than 250 respondents. This method is usually the most �hands on� of the methods listed above for a researcher, as they must define the following three properties for each analysis: (1) how similarity or distance is measured, (2) how clusters are aggregated, and (3) how many clusters are needed. Because of small sample sizes, hierarchical segmentation often results in clusters that are nested instead of being mutually exclusive, meaning that larger clusters created at later stages in the analysis may contain smaller clusters created at earlier stages.

K-means clustering uses Euclidean distance to minimize variance with in each cluster and maximize variance between clusters. The researcher must initially designate the number of clusters he or she wants to develop, and then the data is �passed through� several times with the cluster centers changing on each pass until an appropriate segmentation is reached.

Two-step clustering is often used with categorical variables, and is also recommended for very large sample sizes because it only requires one pass through the data set (whereas both hierarchical and k-means clustering require multiple passes). The method has two steps: first, cases in the data set are grouped into �pre-clusters,� which are then treated as single cases in a hierarchical cluster analysis to reach a final set of segments.

End Uses of Segmentation Analysis

The key to a good segmentation, as with any analysis, is that the results must be projectable to the population of a client�s competitive market. In other words, researchers must be sure that the segments he or she develops can easily translate from a written marketing research report to the client�s actual competitive environment. In the 1978 book Research for Marketing Decisions, authors Paul Green and Donald Tull established the following four basic criteria for a successful segmentation: (1) The segments must exist in the actual competitive environment, (2) the segments must be identifiable (repeatedly and consistently), (3) the segments must be reasonably stable over time, and (4) one must be able to efficiently reach the identified segments.

Provided the four criteria above are met, a segmentation analysis can have many uses in a client�s competitive environment. One of the most common uses is to develop a target market strategy, where companies utilize market segmentation to determine which customer clusters to target, as well as which products or services to offer to those particular clusters. In the same vein, segmentation analysis can be very useful in product positioning, where executives define the image of a product or service they want to project to customers in the clusters they choose to target.

In addition to the above, segmentation has proved useful in designing products and services to meet market needs, gauging a company�s market position (how a company is perceived by its customers relative to its competition), and conducting general fine-tuning to current marketing strategies.

No matter what your industry, segmentation analysis can be a valuable research tool to guide you to an effective marketing strategy for your competitive environment.

All Content � Copyright 2004-2011. Polaris Marketing Research, Inc. All Rights Reserved.

Send inquiries to or call 1-888-816-8700.