Summary

Using GLM Multivariate, you have examined the profiles of churners and non-churners across predefined customer types for various services. The results have given you several things to think about.

• First, across customer types, the amount spent on long distance and equipment services provide good separation between churners and non-churners.
• For Total service customers, the amount spent on toll free service provides some extra separation between churners and non-churners.
• Lastly, for E-service and Basic service customers, the amount spent on calling card services provides some extra separation.

The use of GLM Multivariate with this data file may be considered somewhat questionable, because there numerous 0's for customer spending when a customer does not have a service. Thus, the error terms will almost certainly be non-normal. However, you can think of these results as more "prospective" than inferential, and therefore helpful as a preliminary step to building predictive models for customer churn.