I worked with a Greek statistician who would always try to correct my pronunciation of the Greek letter chi.  I would say “kai” and he would say something similar to “he”.  It was like he and key combined.  I can’t do it justice so I continued to say kai. 

Regardless of how you pronounce it, the chi square test can be very useful.  In fact, one of my business school classes was spent discussing the uses and assumptions of the chi square test.  I won’t try to summarize a semester’s worth of material into a blog post.  Rather, I wanted to point out that chi square tests are used for categorical data and the only “gotcha” is that you have to use the actual counts (rather than percentages).  It is sensitive to cell counts and requires that there be at least 5 observations per cell. 

The chi square test is a powerful statistical tool as it can tell you if there are significant differences between categories and it is the foundation for CHAID.  CHAID is an abbreviation for CHi-square Automated Interaction Detector.  It is one of the many segmentation techniques used in marketing and, if you plot out the tree that results from CHAID, it is a wonderfully visual way to see differences within your customers and/or prospects.  For CHAID you will need to define a dependent variable and undergo EDA (exploratory data analysis) similar to a modeling project.

My clients often know who their best customers are.  Typically the best are the top 20% of customers that generate 80% of the profits.  These are the customers you most want to retain.  The question becomes who are the customers that you should try to migrate into your best customer segment?  Figuring out who are the next best requires research into their behaviors, demographics or firmographics, and attitudes. 

 Segmentation is one way to separate your customer base into differentiated groups against which relevant marketing communicationsand strategies can be developed and executed.  There are many different types of segmentation and techniques including cluster analysis, RFM and CHAID.

Regardless of what method you choose, bear in mind that a good segmentation scheme is often a result of art and science.  Segments should make sense intuitively and, if they are data driven, should be sound statistically.  In my next post I will describe clustering and how that is used for segmentation.

I ran a 5K race recently and so I have been thinking about my pre-race routine.  I may not have a lucky rabbit’s foot but the morning of a race I have a ritual of sorts.  I eat my usual breakfast and of course drink coffee.  There are some things I just can’t do without!  I wear the same clothes and sneakers for the race as I wore training.  I will not do anything new or different.   

This ritual keeps me from being distracted so that I can concentrate on the race.  In this case, my ritual helps me.  But in the office, rituals can be limiting.  Always doing something the same way can get old and stale.  A colleague asked me about identifying best customers.  My first thought was an RFM or RAD segmentation because I was in the midst of a RAD segmentation.  It would have been easy to stop there.  However, I couldn’t stop until I also suggested clusters and CHAID.   If she had let me, I would have added modeling and NPV.   The trick is knowing when to stick with rituals and when to avoid them.