5 Understanding churn and behavior with metrics

 

This chapter covers

  • Showing how churn relates to metrics using cohort analysis
  • Summarizing the range of customer behaviors with dataset statistics
  • Converting metrics from their normal scale to scores
  • Removing invalid observations from a cohort analysis
  • Defining customer segments based on metrics and churn

If you need to use statistics to understand your experiment, then you ought to have done a better experiment.

—Ernest Rutherford, Nobel Prize in Chemistry, 1908, known as “The Father of Nuclear Physics” for his discovery of radioactive decay

It’s time to do what you came here for: understand why your customers are churning and what keeps them engaged. Although it took a while, the dataset you learned to create in chapters 3 and 4 is the foundation for what comes next. You might expect that now I’m going to dive into some serious statistics or machine learning to do the analysis. Instead, I want to call your attention to the quote at the top of the page, which is my favorite saying by a scientist.

5.1 Metric cohort analysis

5.1.1 The idea behind cohort analysis

5.1.2 Cohort analysis with Python

5.1.3 Cohorts of product use

5.1.4 Cohorts of account tenure

5.1.5 Cohort analysis of billing period

5.1.6 Minimum cohort size

5.1.7 Significant and insignificant cohort differences

5.1.8 Metric cohorts with a majority of zero customer metrics

5.1.9 Causality: Are the metrics causing churn?

5.2 Summarizing customer behavior

5.2.1 Understanding the distribution of the metrics

5.2.2 Calculating dataset summary statistics in Python

5.2.3 Screening rare metrics