BrandNewMovers girl picture

Click below to receive 20% OFF your first online order today!

Click here for 20% off your first order!

Sign up for your FREE account today and receive 20% OFF
your first online order!

Thawte certificate

Database Marketing - In Search of Statistical Significance

By Kostis Panayotakis
The goal of database marketing is to increase marketing efficiency & Customer lifetime value, with the smart use of Customer data. In example, use Customer data to identify Customer groups, which would yield high response to offers, in order to address them directly.

Database marketing is based on Customer information related to:
• Customer behavior
• Customer profile & demographics

Based exclusively on behavioral information, one can classify customers into RFM (recency - frequency - monetary) or RF cells. The goal is to identify Customer groups with high expected response rates. Different RFM cells are expected to provide significantly different expected response rate (especially the ones linked to the most recent Customers). The more significant the statistically expected difference is, the higher potential business value this grouping yields. In order to apply RFM, one does not need statistics skills. Therefore this approach is less costly, since it is simpler and requires only customer behavioral information.

Predictive models based on both behavioral & demographic data, can outperform Customer groupings based solely on behavioral data (like RFM).

In order to develop such a model, one needs to use behavioral & demographic data of a set of Customers, which have been monitored vis-à-vis their responses to a specific offer. The Customer set is divided into two subsets of equal (or comparable) size & similar types of Customers (in respect with profile & behavior): a test set (or model train set) and a validation set. The model shall be developed against the test set and validated against the validation set. A modeling algorithm can be applied (e.g. logistic regression analysis), against the test set data, in order to identify the variables, that influence significantly ‘the probability to respond to an offer’ (which is the dependent variable). Validation of the model follows. It involves identifying most of the actual responders in the validation set, given that these Customers are known. After being validated, the model can be used in a test campaign.

Various obstacles may appear during this modelling process:
• There may be no capture of customer reactions to previous offers, therefore no data to model on.
• If the model does not validate sufficiently against the validation group, then the model may be a failure. This may mean that factors affecting significantly the customer behavior, are not captured among the data available or are not used in the model.
• Many customer databases hold Customer behavior info, but limited demographics on the Customers. Lists with consumer demographics (offered by many in the USA), can be used to enrich Customer data with demographics.
A validated model can be applied on the whole Customer database, to identify a group of Customers with high propensity to respond positively to a similar offer. Having produced this Customer list, the next step is to run a test campaign in order to verify the expected response and analyse again the results. Any attempt to execute a fully blown campaign without a prior test, may lead to a failure, since market conditions are constantly changing.

Copyright 2006 - Kostis Panayotakis
Material relevant to database marketing, can be found on

Article Source:

April 8, 2020

Number of new homeowners
added in the past 30 days:


Find New Homeowners Now!

Newsletter logo FREE
Direct Marketing Tips

Sign up for our FREE e-newsletter!