Churn prediction approach

Can anyone please suggest how to build a robust churn prediction model ? I have six months of user subscription data of an internet service provider .

Thank you .

0 members like this

▶ Reply to This

Replies to This Discussion

Permalink Reply by Boyd Owens on May 10, 2016 at 5:46am

One common model is a "Friction" model. The idea is that there is no one single issue that is likely to cause a subscriber to leave and change ISP's, but that a series of events build up over time and eventually prompts the action to move to another ISP. These friction events (Variables) can be things like the

- the number of disconnections in service

- the number of tech support calls

- the number of different types of tech support calls

- the number of truck rolls to the customers home

- the number of failed devices in the home

- Sentiment analysis of text comment in surveys or Tech support cases that show Anger or Frustration.

- Bad followup survey results.

Each of these variables should have a weight based on severity of the events and the model then adds up these weighted events over time to predict churn. The top level friction variable is a measure of the difficulty of continuing to do business with the ISP. This type of model is the type that I have seen used by several large Telco's in the past 5 years.

This type of model has also been used for predicting employee churn.