Customer retention is a critical stage for customer relationship management (CRM), especially for established businesses after their initial exponential growth. Churn management or attrition management is important as when customers leave, there arenegative impacts on revenues. Churn analytics has been widely applied to proactive customer retention where descriptive and predictive analytics are utilised to identify and predict customer propensity to churn.
Saturn Telecommunication is conducting an analysis on their existing customer base with their demographics information, account information and service status recorded. As a business analyst, I analysed the data to provide insights of the churn population and developed as well as evaluated predictive models for customer retention purposes.
The project is seeking insights and solutions relating to: • Understanding the characteristics of its churned and non-churned customers; • Understanding the characteristics of loyal customers (i.e. customers who do not churn and are above a certain threshold of the tenure value*); • Developing and evaluating models to predict customer propensity to churn; • Recommending potential campaigns to buy back or win back the valued customers who churned.
Task 1: Understanding the characteristics of churned, non-churned customers and loyal customers Conducted descriptive analysis based on the customer data and constructed customer profiles for churned, non-churned and loyal customers.
Task 2: Developed and evaluating models to predict propensity to churn. I answered the following questions: a) What is the overall churn rate and the group churn rate for each categorical variable? (For example: senior citizen (yes and no), partner (yes and no), etc.) Identify the categorical variable that has the highest group churn rate. b) Use SAS Enterprise Miner to develop and evaluate at least three predictive models for churn prediction. • Apply standardization (z-score normalization) on the continuous/interval variables. Why you need to apply this? • What are the selected variables used for building the prediction models? • What are the predictive performance of various models and how they rank against one another? • Discuss which is the best model and how do you best interpret the model?
Task 3: Campaign recommendations based on insights obtained from Task 1 and Task 2 Provide campaign recommendations based on insights obtained from the first two tasks above.