How do you define membership or subscription churn?
Here is a rather technical article that describes a way of thinking about the subject.
WTTE-RNN - Less hacky churn prediction
Please share how you describe churn and if you make it through this article if you are thinking about churn and churn prediction differently.
--Tom
Hi Tom, This is a neat article, thanks for sharing. Churn is one of those things that seems deceptively simple and this article shows just how intricate it can be. At TLCC we're doing a roadmap session collecting use cases for a Constituent-based "behaviors" cube. I think churn reporting is a great use case for a data model focused on user interactions.
Looking forward to the conversation.
We potentially have a lot of good behavioral data in Tessitura to do this kind of modeling.
Orders, memberships, subscriptions, promotion records as a few examples.
As this article points out, churn modeling is actually a form of time series analysis. Having data “As of” a particular event in the past is important to churn modeling (or survival analysis).
In the work that I have done, things like postal code “as of” the beginning and end of the membership provide some useful insights. Right now this type of analysis is not very easy because of current data storage techniques, bugs that exist(ed) in the system making this data incomplete, and more difficult to access than one might hope.
John Jakovich have you had any further thoughts about how to make this kind “as of” analysis easier for the community? So that we can make more progress with Data Science and ML?