Customer Segmentation - Data Mining Techniques in CRM: Inside Customer Segmentation - page 212

• Clean your data from obvious segments (e.g., inactive customers) before pro-
ceeding with the segmentation analysis.
• Always bear in mind that eventually the resulting model will be deployed. In
other words, it will be used for scoring customers and for supporting specific
marketing actions. Thus, when it comes to selecting the population to be used
for model training, do not forget that this is the same population that will be
scored and included in a marketing activity at the end. So sometimes it is better
to start with the end in mind and consider who we want to score, segment,
or classify at the end: the entire customer base, consumer customers, only
high-value customers, and so on. This deployment-based approach can help us
to resolve ambiguities about selection of the modeling dataset population.
• Select only variables relevant to the specific business objective and the particular
behavioral aspects you want to investigate. Avoid mixing all available inputs in an
attempt to build a ''magic'' segmentation that will cover all aspects of a customer's
relationship with the organization (e.g., phone usage and payment behavior).
• Avoid using demographic variables in a behavioral segmentation project. Mixing
behavioral and demographical information may result in unclear and ambiguous
behavioral segments since two customers with identical demographic profiles
may have completely different behaviors.
• Consider the case of a father who has activated a mobile phone line for his
teenage son. In a behavioral segmentation solution, based only on behavioral
data, this line would most likely be assigned to the ''Young - SMS users''
segment, along with other teenagers and young technophile users. Therefore
we might expect some ambiguities when trying to examine the demographic
profile of the segments. In fact, this hypothetical example also outlines why
the use of demographic inputs should be avoided when the main objective is
behavioral separation.
• 'Smooth' your data. Prefer to use monthly averages, percentages, ratios, and
other summarizing KPIs that are based on more than one month of data.
• A general recommendation on the time frame of the behavioral data to be used
is to avoid using less than 6 months and more than 12 months of data in order
to avoid founding the segments on unstable/volatile or outdated behaviors.
• Try different combinations of input fields and explore different models and
model settings. Build numerous solutions and pick the one that best addresses
the business goal.
• Labeling the segments needs extra care. Bear in mind that this label will
characterize the segments, so a hasty naming will misguide all recipients/users
of this information. A plain label will unavoidably be a kind of derogation, as it is
impossible to incorporate all the differentiating characteristics of a segment. Yet,
a carefully chosen name may simply and successfully communicate the unique
characteristics of the segments to all subsequent users.