The Art and Science of Constructing Powerful Predictors


The Art and Science of Constructing Powerful Predictors

Presented at the NCDM Orlando ’99 Conference and Exhibition by Perry D. Drake and Rhonda Knehans-Drake


Perry D. Drake and Rhonda Knehans-Drake presented various techniques used to manipulate and exploit customer data to attendees of the NCDM Conference and Exhibition (Orlando, December 7, 1999). In addition, they shared various methods a direct marketer can use to become intimate with the customer data and the importance of this step prior to analyzing the data.

Toimages Discussed included:

Getting to Know Your Customer Data
Without full comprehension of customer data, how can it be exploited to its fullest potential?

Various Data Manipulation Techniques
1. Univariate Analysis
Viewing one data element at a time to determine its strength in terms of helping predict the customer action of concern, such as order or payment. (The most utilized analysis technique by direct marketers.)

2. Cross-tabulation Analysis
Providing a means of viewing two or more data elements in combination and allows one to highlight any interrelationships among variables.

3. Combined Logic Terms
Providing the direct marketer a method to combine several related predictors into a single, more powerful predictor of customer behavior.

4. Ratio Variables
Dividing one continuous data element by another has the ability to turn relatively weak predictors into very powerful predictors.

5. Longitudinal or Time Series Variable Analysis
Providing a view of a particular data element for each customer across time. A useful tool in “life-stage” marketing efforts.

6. Time Alignment of Key Events for Continuous Marketing Service
Allowing for a better assessment of customers enrolled in continuous services such as clubs, continuities, and frequent buyer programs. Without proper time-alignment, assessment of customers is misleading.

7. Driving In-House Data Down to a Geo-Demographic Level
Driving your own house data to a geo-demographic level (zip-code, block groups, etc.) allows you to enhance data poor lists or outside lists with valuable information aiding in the selection of names for promotional consideration.