6 Relationships between customer behaviors

 

This chapter covers

  • Analyzing relationships between pairs of metrics
  • Calculating matrices of correlation coefficients
  • Calculating averages of correlated metric scores
  • Segmenting customers using averages of metrics
  • Discovering metric groups with clustering

For most products and services, analyzing whether individual metrics are related to churn is the beginning but not the end of using your data to reduce churn. This chapter teaches you how to address a common problem: having an overabundance of data available for fighting churn. In the age of big data, some companies collect a lot about their customers. That should make it easier to fight churn with data, right? Not quite.

Many customer behaviors are closely related, so metrics based on those behaviors have similar relationships to churn. A cohort churn analysis on a typical company’s database of events and metrics probably won’t give you just a few cohort churn plots: you probably have dozens or more. This can actually cause more confusion than good. When behaviors measured by metrics are not the specific acts that give enjoyment or utility to the user, then the relationships to churn are just associations and not causal. When you have a lot of metrics that are associated with churn but not causal, you don’t have a good way to understand how they act together.

6.1 Correlation between behaviors

6.1.1 Correlation between pairs of metrics

6.1.2 Investigating correlations with Python

6.1.3 Understanding correlations between sets of metrics with correlation matrices

6.1.4 Case study correlation matrices

6.1.5 Calculating correlation matrices in Python

6.2 Averaging groups of behavioral metrics

6.2.1 Why you average correlated metric scores

6.2.2 Averaging scores with a matrix of weights (loading matrix)

6.2.3 Case study for loading matrices

6.2.4 Applying a loading matrix in Python

6.2.5 Churn cohort analysis on metric group average scores