|
The Same Cluster Exercise on Internal Data
As market research is only carried out on a small
sample of your customer base, you only know actual
segment membership for this small section. How can
the thousands of other customers be allocated to
segments? Market research gives you an important
insight into your customers and differentiates
different segments, but have you put this market
research segmentation into practice? Can you find
the same customer segments from the information in
your databases?
The Challenge
On one hand, it is very difficult or often even
impossible to use predictive techniques to find
models that are accurate enough for predicting
segment membership. On the other hand, creating a
new segmentation based on only internal data will
rarely lead to groups that correspond with the
descriptions of the segments found in the market
research segmentation.
IKAN Consulting Solution
2 years ago, IKAN Consulting worked out a solution
for overcoming most of these problems and for being
able to create a segmentation that can be put into
practice and that is acceptable for market research
and business users. It has now implemented this at
several companies. First we receive the different
segment definitions developed by the market research
department. Then we cluster using the company’s
internal transaction data, in order to find these
segments in the customer database.
Bayesian Expectation Maximization Clustering
We have developed an enhancement to the EM
clustering algorithm, which we have called the
Bayesian Expectation Maximization Clustering. This
algorithm uses the market research clustering as a
target. The result we find is a stable and coherent
clustering solution in the database, which defines
segments that respect the core characteristics of
the market research segmentation in terms of
attitude, needs, values and/or actual behavior.
Because of the number of different segments and the
settings of the initial seeds for each individual
segment, based on the a priori knowledge of the
market research clustering, the BEMC algorithm
allows you to control the sizes of the individual
segments and direct each cluster using one or more
variables. It also creates individual probabilities
of belonging to a certain segment. Generating
probabilities instead of the sole segment allocation
is important for the use of the segmentation in
operational marketing.
The IKAN Consulting Data Mining Team
Our team is made up from several members with
various backgrounds, from experienced marketers to
statisticians and engineers to mathematicians. As
well as combining market research data and internal
data, we also have a proven track record in Churn
Prediction, Cross and Up Selling, Expected Lifetime
Value, Social Network Analysis and Sales Prediction.
You name it, we’ve done it. Data mining is our
passion and our specialty.
|
