Post-hoc segmentation often utilizes information collected via surveys. If properly designed, these surveys can help to classify respondents into groups that are relatively homogenous and distinct from one another. For many businesses, existing data sources can also be used to segment groups, such as transaction records (which include variables such as company size, the amount of business conducted, and geographic indicators, among others.)
Segmentation models also can be derived from a variety of input or "basis" variables. Some examples of basis variables include product attribute preferences, benefits sought, product usage profiles, and price sensitivity, among many others. The key is to identify a reasonable number of groups of respondents who provide similar answers to the basis variable questions. These questions must be carefully considered to ensure that the results of the segmentation provides an accurate market model that identifies target customers or clients, but also clearly identifies distinct needs, wants, and behavior patterns.
Types of Post-Hoc Segmentation
There are three primary methods used in 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 requires that the researcher or analyst determine three properties that will be used to analyze the data, including how similarity or distance is measured, how the data clusters are aggregated, and the number of clusters that are needed. Data must be crunched several times in order to ensure that each cluster is as distinct as possible, and different enough from the other clusters. However, given the relatively small sample size, some clusters may be nested inside of others, rather than being mutually exclusive.
For larger sample sizes, analysts can use k-means clustering, which involves the use of Euclidean distance to establish a minimum variance within each cluster (ensuring that the cluster is, in fact, a distinct grouping of respondents, with similar characteristics, views, or behaviors.) K-means clustering also ensures that each cluster is distinct from all others, to prevent clusters from overlapping each other, and muddying the resulting analysis. In k-means clustering, the researcher or analyst must initially designate the number of clusters desired, and then the data is crunched in successive passes, with the cluster centers changing on each pass, until the desired segmentation is achieved.
Finally, for very large sample sizes, two-step clustering is often deployed because it only requires a single pass through the data set in order to provide results. Two-step clustering requires that cases (individual observations, participants, or survey respondents, etc.) are grouped together into pre-clusters. Then, these pre-clusters are then treated as single cases in a hierarchical cluster analysis to reach a final segmentation.
Regardless of the method used, post-hoc segmentation analysis can yield a significantly greater level of detail and insight into customer or client behaviors, and allow a far more precise strategy for micro-marketing and targeting to be developed.