Practical Problem Solving - Java Data Mining: Strategy, Standard, and Practice - page 314

12.3 Business Scenario 3: Using Customer Segmentation

HEW has been successfully using the previous applications to
optimize direct marketing campaigns. The chief marketing officer
(CMO) now wants to advertise in select mass media. For this, he
needs to know more about the customer profiles to target the proper
magazines and TV channels.
The marketing term for getting these profiles is segmentation . Cus-
tomer segmentation can be done in many ways, and those familiar
with the CRM space know that the subject of customer segmentation
could fill a topic of its own; in fact, many topics have been written on
the subject (see, for example, Optimal Database Marketing: Strategy,
Development and Data Mining [Drozenko/Drake 2002]). Most cus-
tomer segmentation efforts are designed to optimize promotional
product offerings and are generally made in several stages. First,
customer segments are made based on information available in the
customer database, such as customers engaged in fraud, a “do-not-
promote” segment of those who do not want to be contacted
anymore, and high risk accounts to be discarded from the next mass
mailings.
Then, the remaining population can be segmented using what is
called life-stage segmentation , which is primarily based on demo-
graphic and psychographic data obtained internally or externally.
These segments can be used to determine the future needs of HEW
customers or to adapt promotions or messages to customer interests.

12.3.1 Customer Segmentation Specifications

To perform the customer segmentation operation, the CMO will buy
psychographic data, such as customer hobby, music, movie, or book
preferences, from an external service bureau or data enhancement
facility. In this example, the service bureau is called “The Service
Bureau” (TSB). This company can augment a file containing names
and addresses with both demographic information, such as house-
hold size, household income, the age of the head of household in
years, and psychographic information, such as interests and hobbies.
This data will be used to create a small number of clusters
(between 5 and 10) for the entire HEW customer database. Then,
these clusters will be characterized in terms of important demo-
graphic and psychographic attributes and each cluster will manually
be assigned a descriptive name. Recognizing a good clustering from