Every data set is unique, whether, it is pure research data or the result of a data fusion between research, internal and geodemographic tagging. No one can tell in advance which technique will work best.
Segmentation is an iterative process with alternative solutions. The aim is to identify homogeneous sub-groups that have different needs and behaviors, but, these need to be useful and identifiable if they are to be fully utilized.
Typical segmentation techniques include k-means (scale data), latent class (mixed data) and two-step (mixed data).
Output includes profiles and a classification algorithm.
Classification techniques also have strengths and weaknesses. Proven ones include discriminant analysis, multinomial discriminant analysis, classification and regression trees (e.g CHAID). Emerging ones that require client machine learning skills include ‘random forests’.
Output is a classification algorithm, potentially Excel based. Or ‘R’ if this can be implemented in a production level (live) environment.