The first published empirical study of churn prediction in scholarly publishing, using a 6.5-year dataset from an unnamed major academic publisher. The model predicts customer defection from resampled usage patterns — volume and frequency of content downloads — rather than granular behavioural data.
Benefits & outcomes
Ensemble models forecast churn nearly a year in advance with substantial accuracy.
Performance held even with behavioural attributes stripped out: simple temporal usage data is enough.
Notes & quotes
"Modelling churn on the basis of resampling customers’ use of resources over subscription time is a better (simplified) approach than when considering the high granularity that can often characterise consumption behaviour."
Publisher / company: Undisclosed Major Academic Publisher
Date of mention: 18/11/2022
Source: https://arxiv.org/pdf/2211.09970