Discover the Joy of Data Segmentation

If you're a database marketer, you should derive great pleasure from the discovery phase of segmentation.

Yes, segmentation should be fun. Too often we view the art of segmentation as a cold science. We follow best practices, using elements of RFM and channel to identify high-spending customers.

For me, the most enjoyable element of segmentation is the discovery phase. Here, there are no rules, no best practices.

I start by creating a data set that has many unique attributes. This is the place to be creative! I define recency by days since last purchase (not weeks, months or years), for instance. I define monetary as both a dollar amount and an average order value. I create dollar fields by year, amount spent in the past 12 months, amount spent 13 months to 24 months ago, and so on.

I include any demographic or lifestyle variables I have, like age or income.

I also add fields to the database that summarize the amount spent in different merchandise classifications. Certain merchandise classifications tend to yield customers that generate more profit over time, so you want to make sure to identify the merchandise classifications that push the brand forward.

Once I have all of my variables defined as I see fit, I use Chi-squared Automatic Interaction Detector (CHAID) as my discovery methodology of choice. I like to let the software methodology pick segments up to a depth of 10 levels. This style of analysis allows me to see interactions in the data. For some customers, RFM quickly rises to the top, while for other customers, merchandise categories and first-purchase attributes make a big difference.

I have two goals in mind with this analysis: One, I want to identify a good segmentation strategy. CHAID helps me find a strategy that'll work reasonably well.