I have to built a customer churn model for of a teleco. The churn rate is 15 %. There is no particular campaign conducted. By churn I mean customer leaving the teleco permanently.

Data is available from Jan 2012 to Dec 2014


  1. Should i have a fixed performance window? i.e. i will only flag customers who churned from July 2014 to Dec 2014 as churners & use only data from before July 2014 to create predictors in the model.
  2. I will consider customers who churned from July 2014 to Dec 2014 only but performance window will be rolling. Example if a customer churned in nov 2014 i will create predictors using data before nov 2014. If a customer churned in sept 2014 i will create predictors using data before sept 2014.

Which is the best approach, using fixed performance window or rolling performance window? What is the pros & cons of these 2 approaches?


  • $\begingroup$ It would be helpful to know whether the act of churning is factual, i.e. in case of breaking of the subscription contract it is, but if we are talking about pay-as-you-go, than it's not. $\endgroup$
    – coulminer
    Feb 17, 2015 at 15:24
  • 1
    $\begingroup$ This for businesses to business. So customers break the contract when they churn. $\endgroup$ Feb 17, 2015 at 15:58
  • $\begingroup$ @ML_Pro, My recommendation is to use a rolling window. This will help you incorporate seasonality (if it exists) in your model, and it will make your model more generalizable. I can elaborate on this, if you're still looking for a detailed answer. $\endgroup$
    – Vishal
    Feb 9, 2018 at 19:33
  • $\begingroup$ Yes. Can you elaborate as an answer pls $\endgroup$ Feb 10, 2018 at 8:52

1 Answer 1


It will depend upon

  1. Provisioning period required for concerned team to act upon churn report generated by model
  2. Complete set data available at start of the month for you to score the customers. For example, if you are scoring customers for month of nov-2014 using predictors till oct-2014 then it only gives 30 days to do customer scoring and running campaign

If it is a b2b model, then go for rolling prediction window as usage patterns might change from month to month which can help model to capture variance in data but you should try rolling prediction window period of 2 or 3 months instead and choose which is better in terms of performance and also gives enough time to act upon


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.