This section outlines the most common methods for segmentation and targeting. It assumes that you have already obtained the data for segmentation (data on basis variables) and, optionally, the data for targeting. These data should be first assembled into usable matrices. A separate technical note describes methods for behavior-based segmentation using choice models.
Broadly stated, there are two approaches to segmentation (Wedel and Kamakura 2000), namely, a priori methods and post-hoc methods. In a priori methods, an analyst uses domain knowledge to segment customers into different groups (e.g., male and female customers). We will not be focusing on these types of approaches here. In post-hoc methods, the analyst relies on data analysis to identify "groupings" of customers.
There are two broad categories of post-hoc methods: (1) Traditional methods, which are based on using a distance or a similarity metric to determine how far or near a customer is from other customers in the market, and (2) Newer probability-based, such as latent cluster anaysis, which can help identify groupings in the population from which a sample of respondents has been selected for the segmentation analysis. If the latent class method results in a well-partitioned segmentation scheme, then it means that each customer in the market belongs to just one segment with high probability.
There are also two broad categories of methods available for targeting analysis, which can be used after we determine the number of segments in the market: (1) Scoring methods, such as discriminant analysis, which can be used to compute a unique score for each customer or prospect. Based on their discriminant scores, customers can be assigned to one (or, sometimes, more than one) of the identified segments. (2) Tree based methods, such as CART (Classification and Regression Trees) and CHAID (Chi-Squared Automatic Interaction Detector).
Contents
Introduction
Traditional Cluster Analysis
- Data and Variable Refinements
- Choosing the set of variables for analysis:
Using factor analysis to reduce the data
Specifying a Measure of Similarity
- Similarity-type measures
Distance-type measures
Segment Formation
- Hierarchical Methods
Partitioning Methods
Interpreting Segmentation Study Results
- How many clusters to retain?
How good are the clusters?
Segment Formation Using Latent Cluster Analysis
- Outline of Latent Cluster Analysis
Model Estimation
Interpreting and Using Results From Latent Cluster Analysis
- Log-Likelihood values
AIC and BIC criterion
Cross entropy
Profiling and Targeting Tools
- Discriminant Analysis
- Interpreting Discriminant Analysis results
Classification Trees
- Outline of classification tree methods
Summary
References
Download
PDF Ebook Segmentation and Targeting: Marketing Engineering Technical Note
