In the previous section I had described the process of using Hotspot profiling in customer segmentation and used a medical insurance example to describe the process.

Before detailing the Hotspot profiling method, let us see what the same example would yield when put through the conventional cluster analysis process of segmentation.

Cluster Analysis

Cluster Analysis groups individuals within a population based on a number of parameters across dimensions including relationship between individuals or benchmarks such as age, occupation and occupational hazards.

A simple two-dimensional example depicting medical insurance can be used with each point representing a consumer.

if we plot a graph with age on the 'y' axis and probability of claim on 'x' there are two clear groupings of consumers in terms of age. Now these groupings may or may not be significantly different in terms of clearly separate market groupings. To make the analysis more realistic and deductive third, fourth or fifth dimensions can be added. But searching for clusters in multi-dimensional space is time consuming and the sheer complexity of the graph makes it an unfeasible operation. Therefore computers in recent times have enabled such extremely complex taxonomic procedures to be performed on large fields of data with multiple objects (members in our case) and variables.

The advantage of using multi-dimensional analysis is that, unlike other methods, it does not rely on a series of dichotomies using a single segmentation criterion at a time. Rather it considers the population simultaneously in all the stated dimensions.

Deciphering Cluster Analysis - An Example

Let us extend one aspect of the above example to identify the target audience (from the existing population) for a new health insurance plan.

Several small groups of individuals are subjected to an attitudinal questionnaire in order to isolate what seems to be the major personal constructs relating to the plan membership. Demographic and socio-economic data of the population in need of the health plan may also be considered.

This way we have elucidated as many as a dozen relevant dimensions by which the population can be adjudged. Now we can identify clusters of interest and subject them to more stringent criteria by adding more dimensions such as their propensity to enrol into such a health plan.

Using cluster analysis we can make a comparison of these prospective customers. This way we do not rely on the discriminating power of just one or two dimensions at a time, but are allowing natural groupings to form. However, this may result in groupings that are difficult to ‘name’. For instance using a dichotomist criterion such as sex could clearly segment the population into ‘Male’ and ‘Female’. But cluster analysis does not consider such criteria and groupings may end up having a mix of male and female respondents that makes it difficult to describe the group in terms of attitude and scores. This is overcome by using methods like Hotspot profiling which look at the highest probability of an event occurring.

Dr Vikram Venkateswaran

Senior Business Consultant

Covansys- A CSC Company