Customer churn analysis
Customer acquisition is expensive. Understanding which customers are likely to churn and why allows product managers to identify potential pain points and friction within the digital experience that may be contributing to customer churn and for customer success to work to retain those customers most at risk of churn.
Overview
Based on a customers product and website usage, along with data about the customer, predict which customers are likely to churn and identify the key factors that are driving this churn.
TARGET
Will a customer churn in a given time period?
Challenge
The first challenge in churn prediction is defining what the business means by churn.
Action
- Cancel a subscription
- Last purchase more than a month or quarter ago
- Last used the product a month or quarter ago
Time Frame
- In the next month, quarter or year.
The next challenge is joining and transforming the data contained in multiple different tables to create the training dataset. This includes data that defines if a customer has churned from the CRM, data that describes how a customer interacts with the product from the product usage and web analytics DB, and enrichment data such as demographics for third party customer behavior and economics data for macro socio-economic influences. All this data needs to be cleaned, joined and transformed into valuable ML features before going into model training. This pre-modeling prep process can be frustrating and time consuming. We are here to help.
Modeling techniques and libraries
Multiple regression analysis
Allow more than the two choices across multiple dimensions. Models the target as a function of the independent variables.
Package:
- Sklearn - linear_model
- Statsmodels
Machine learning analysis
Build machine learning models to predict the target as a function of the independent variables. Use model interpretability packages to evaluate the impact of the independent variables on the prediction.
Packages:
- Sklearn
- ELI5
- LIME
- SHAP
