Customer Segmentation - Data Mining Techniques in CRM: Inside Customer Segmentation - page 208

The most widely used criterion for deciding the number of components
to extract is ''Eigenvalues over 1.'' However, the final decision should also take
into account the percentage of the total variance explained by the extracted
components. This percentage should, in no case, be lower than 60-65%.
Most importantly, the final components retained should be interpretable
and useful from a business perspective. If an extra component makes sense
and provides conceptual clarity then it should be considered for retention, as
opposed to one that makes only ''statistical'' sense and adds nothing in terms
of business value.
Identification of the Segments with Cluster Modeling
Customers are divided into distinct segments by using cluster analysis. The
clustering fields, typically the component scores, are fed as inputs into a cluster
model which assesses the similarities between the records/customers and suggests a
way of grouping them. Data miners should try a test approach and explore different
combinations of inputs, different models, and model settings before selecting the
final segmentation scheme.
Different clustering models will most likely produce different segments and
this should not come as a surprise. Expecting a unique and definitive solution
is a sure recipe for disappointment. Usually the results of different algorithms
are not identical but similar. They seem to converge to some common segments.
Analysts should evaluate the agreement level of the different models and examine
which aspects disagree. In general, a high agreement level between many different
cluster models is a good sign for the existence of discernible groupings.
The modeling results should be evaluated before selecting the segmenta-
tion scheme to be deployed. This takes us to the next stage of the behavioral
segmentation procedure.
Evaluation and Profiling of the Revealed Segments
In this phase the modeling results are evaluated and the segmentation scheme
that best addresses the needs of the organization is selected for deployment. Data
miners should not blindly trust the solution suggested by one algorithm. They
should explore different solutions and always seek guidance from the marketers
for selecting the most effective segmentation. After all, they are the ones who
will use the results for segmented marketing and their opinion on the future
benefits of each solution is critical. The selected solution should provide distinct
and meaningful clusters that can indicate profitable opportunities. Tasks in this
phase include: