Fighting Churn with Data

Fighting Churn with Data
brief contents
contents
foreword
preface
acknowledgments
about this book
	Who should read this book
	How this book is organized: A road map
	About the code
	liveBook discussion forum
	Other online resources
about the author
about the cover illustration
Part 1—Building your arsenal
	1 The world of churn
		1.1 Why you are reading this book
			1.1.1 The typical churn scenario
			1.1.2 What this book is about
		1.2 Fighting churn
			1.2.1 Interventions that reduce churn
			1.2.2 Why churn is hard to fight
			1.2.3 Great customer metrics: Weapons in the fight against churn
		1.3 Why this book is different
			1.3.1 Practical and in-depth
			1.3.2 Simulated case study
		1.4 Products with recurring user interactions
			1.4.1 Paid consumer products
			1.4.2 Business-to-business services
			1.4.3 Ad-supported media and apps
			1.4.4 Consumer feed subscriptions
			1.4.5 Freemium business models
			1.4.6 In-app purchase models
		1.5 Nonsubscription churn scenarios
			1.5.1 Inactivity as churn
			1.5.2 Free trial conversion
			1.5.3 Upsell/down sell
			1.5.4 Other yes/no (binary) customer predictions
			1.5.5 Customer activity predictions
			1.5.6 Use cases that are not like churn
		1.6 Customer behavior data
			1.6.1 Customer events in common product categories
			1.6.2 The most important events
		1.7 Case studies in fighting churn
			1.7.1 Klipfolio
			1.7.2 Broadly
			1.7.3 Versature
			1.7.4 Social network simulation
		1.8 Case studies in great customer metrics
			1.8.1 Utilization
			1.8.2 Success rates
			1.8.3 Unit cost
		Summary
	2 Measuring churn
		2.1 Definition of the churn rate
			2.1.1 Calculating the churn rate and retention rate
			2.1.2 The relationship between churn rate and retention rate
		2.2 Subscription databases
		2.3 Basic churn calculation: Net retention
			2.3.1 Net retention calculation
			2.3.2 SQL net retention calculation
			2.3.3 Interpreting net retention
		2.4 Standard account-based churn
			2.4.1 Standard churn rate definition
			2.4.2 Outer joins for churn calculation
			2.4.3 Standard churn calculation with SQL
			2.4.4 When to use the standard churn rate
		2.5 Activity (event-based) churn for nonsubscription products
			2.5.1 Defining an active account and churn from events
			2.5.2 Activity churn calculations with SQL
		2.6 Advanced churn: Monthly recurring revenue (MRR) churn
			2.6.1 MRR churn definition and calculation
			2.6.2 MRR churn calculation with SQL
			2.6.3 MRR churn vs. account churn vs. net (retention) churn
		2.7 Churn rate measurement conversion
			2.7.1 Survivor analysis (advanced)
			2.7.2 Churn rate conversions
			2.7.3 Converting any churn measurement window in SQL
			2.7.4 Picking the churn measurement window
			2.7.5 Seasonality and churn rates
		Summary
	3 Measuring customers
		3.1 From events to metrics
		3.2 Event data warehouse schema
		3.3 Counting events in one time period
		3.4 Details of metric period definitions
			3.4.1 Weekly behavioral cycles
			3.4.2 Timestamps for metric measurements
		3.5 Making measurements at different points in time
			3.5.1 Overlapping measurement windows
			3.5.2 Timing metric measurements
			3.5.3 Saving metric measurements
			3.5.4 Saving metrics for the simulation examples
		3.6 Measuring totals and averages of event properties
		3.7 Metric quality assurance
			3.7.1 Testing how metrics change over time
			3.7.2 Metric quality assurance (QA) case studies
			3.7.3 Checking how many accounts receive metrics
		3.8 Event QA
			3.8.1 Checking how events change over time
			3.8.2 Checking events per account
		3.9 Selecting the measurement period for behavioral measurements
		3.10 Measuring account tenure
			3.10.1 Account tenure definition
			3.10.2 Recursive table expressions for account tenure
			3.10.3 Account tenure SQL program
		3.11 Measuring MRR and other subscription metrics
			3.11.1 Calculating MRR as a metric
			3.11.2 Subscriptions for specific amounts
			3.11.3 Calculating subscription unit quantities as metrics
			3.11.4 Calculating the billing period as a metric
		Summary
	4 Observing renewal and churn
		4.1 Introduction to datasets
		4.2 How to observe customers
			4.2.1 Observation lead time
			4.2.2 Observing sequences of renewals and a churn
			4.2.3 Overview of creating a dataset from subscriptions
		4.3 Identifying active periods from subscriptions
			4.3.1 Active periods
			4.3.2 Schema for storing active periods
			4.3.3 Finding active periods that are ongoing
			4.3.4 Finding active periods ending in churn
		4.4 Identifying active periods for nonsubscription products
			4.4.1 Active period definition
			4.4.2 Process for forming datasets from events
			4.4.3 SQL for calculating active weeks
		4.5 Picking observation dates
			4.5.1 Balancing churn and nonchurn observations
			4.5.2 Observation date-picking algorithm
			4.5.3 Observation date SQL program
		4.6 Exporting a churn dataset
			4.6.1 Dataset creation SQL program
		4.7 Exporting the current customers for segmentation
			4.7.1 Selecting active accounts and metrics
			4.7.2 Segmenting customers by their metrics
		Summary
Part 2—Waging the war
	5 Understanding churn and behavior with metrics
		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
			5.2.4 Involving the business in data quality assurance
		5.3 Scoring metrics
			5.3.1 The idea behind metric scores
			5.3.2 The metric score algorithm
			5.3.3 Calculating metric scores in Python
			5.3.4 Cohort analysis with scored metrics
			5.3.5 Cohort analysis of monthly recurring revenue
		5.4 Removing unwanted or invalid observations
			5.4.1 Removing nonpaying customers from churn analysis
			5.4.2 Removing observations based on metric thresholds in Python
			5.4.3 Removing zero measurements from rare metric analyses
			5.4.4 Disengaging behaviors: Metrics associated with increasing churn
		5.5 Segmenting customers by using cohort analysis
			5.5.1 Segmenting process
			5.5.2 Choosing segment criteria
		Summary
	6 Relationships between customer behaviors
		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
		6.3 Discovering groups of correlated metrics
			6.3.1 Grouping metrics by clustering correlations
			6.3.2 Clustering correlations in Python
			6.3.3 Loading matrix weights that make the average of scores a score
			6.3.4 Running the metric grouping and grouped cohort analysis listings
			6.3.5 Picking the correlation threshold for clustering
		6.4 Explaining correlated metric groups to businesspeople
		Summary
	7 Segmenting customers with advanced metrics
		7.1 Ratio metrics
			7.1.1 When to use ratio metrics and why
			7.1.2 How to calculate ratio metrics
			7.1.3 Ratio metric case study examples
			7.1.4 Additional ratio metrics for the simulated social network
		7.2 Percentage of total metrics
			7.2.1 Calculating percentage of total metrics
			7.2.2 Percentage of total metric case study with two metrics
			7.2.3 Percentage of total metrics case study with multiple metrics
		7.3 Metrics that measure change
			7.3.1 Measuring change in the level of activity
			7.3.2 Scores for metrics with extreme outliers (fat tails)
			7.3.3 Measuring the time since the last activity
		7.4 Scaling metric time periods
			7.4.1 Scaling longer metrics to shorter quoting periods
			7.4.2 Estimating metrics for new accounts
		7.5 User metrics
			7.5.1 Measuring active users
			7.5.2 Active user metrics
		7.6 Which ratios to use
			7.6.1 Why use ratios, and what else is there?
			7.6.2 Which ratios to use?
		Summary
Part 3—Special weapons and tactics
	8 Forecasting churn
		8.1 Forecasting churn with a model
			8.1.1 Probability forecasts with a model
			8.1.2 Engagement and retention probability
			8.1.3 Engagement and customer behavior
			8.1.4 An offset matches observed churn rates to the S curve
			8.1.5 The logistic regression probability calculation
		8.2 Reviewing data preparation
		8.3 Fitting a churn model
			8.3.1 Results of logistic regression
			8.3.2 Logistic regression code
			8.3.3 Explaining logistic regression results
			8.3.4 Logistic regression case study
			8.3.5 Calibration and historical churn probabilities
		8.4 Forecasting churn probabilities
			8.4.1 Preparing the current customer dataset for forecasting
			8.4.2 Preparing the current customer data for segmenting
			8.4.3 Forecasting with a saved model
			8.4.4 Forecasting case studies
			8.4.5 Forecast calibration and forecast drift
		8.5 Pitfalls of churn forecasting
			8.5.1 Correlated metrics
			8.5.2 Outliers
		8.6 Customer lifetime value
			8.6.1 The meaning(s) of CLV
			8.6.2 From churn to expected customer lifetime
			8.6.3 CLV formulas
		Summary
	9 Forecast accuracy and machine learning
		9.1 Measuring the accuracy of churn forecasts
			9.1.1 Why you don’t use the standard accuracy measurement for churn
			9.1.2 Measuring churn forecast accuracy with the AUC
			9.1.3 Measuring churn forecast accuracy with the lift
		9.2 Historical accuracy simulation: Backtesting
			9.2.1 What and why of backtesting
			9.2.2 Backtesting code
			9.2.3 Backtesting considerations and pitfalls
		9.3 The regression control parameter
			9.3.1 Controlling the strength and number of regression weights
			9.3.2 Regression with the control parameter
		9.4 Picking the regression parameter by testing (cross-validation)
			9.4.1 Cross-validation
			9.4.2 Cross-validation code
			9.4.3 Regression cross-validation case studies
		9.5 Forecasting churn risk with machine learning
			9.5.1 The XGBoost learning model
			9.5.2 XGBoost cross-validation
			9.5.3 Comparison of XGBoost accuracy to regression
			9.5.4 Comparison of advanced and basic metrics
		9.6 Segmenting customers with machine learning forecasts
		Summary
	10 Churn demographics and firmographics
		10.1 Demographic and firmographic datasets
			10.1.1 Types of demographic and firmographic data
			10.1.2 Account data model for the social network simulation
			10.1.3 Demographic dataset SQL
		10.2 Churn cohorts with demographic and firmographic categories
			10.2.1 Churn rate cohorts for demographic categories
			10.2.2 Churn rate confidence intervals
			10.2.3 Comparing demographic cohorts with confidence intervals
		10.3 Grouping demographic categories
			10.3.1 Representing groups with a mapping dictionary
			10.3.2 Cohort analysis with grouped categories
			10.3.3 Designing category groups
		10.4 Churn analysis for date- and numeric-based demographics
		10.5 Churn forecasting with demographic data
			10.5.1 Converting text fields to dummy variables
			10.5.2 Forecasting churn with categorical dummy variables alone
			10.5.3 Combining dummy variables with numeric data
			10.5.4 Forecasting churn with demographic and metrics combined
		10.6 Segmenting current customers with demographic data
		Summary
	11 Leading the fight against churn
		11.1 Planning your own fight against churn
			11.1.1 Data processing and analysis checklist
			11.1.2 Communication to the business checklist
		11.2 Running the book listings on your own data
			11.2.1 Loading your data into this book’s data schema
			11.2.2 Running the listings on your own data
		11.3 Porting this book’s listings to different environments
			11.3.1 Porting the SQL listings
			11.3.2 Porting the Python listings
		11.4 Learning more and keeping in touch
			11.4.1 Author’s blog site and social media
			11.4.2 Sources for churn benchmark information
			11.4.3 Other sources of information about churn
			11.4.4 Products that help with churn
		Summary
index
	A
	B
	C
	D
	E
	F
	G
	H
	I
	J
	K
	L
	M
	N
	O
	P
	Q
	R
	S
	T
	U
	V
	W
	X