How Early Is Early Enough? Design-Dependent Observation-Window Sufficiency in Subscription Churn Prediction
2026-07-01 • Machine Learning
Machine Learning
AI summaryⓘ
The authors studied how many days of early user behavior are needed to predict if someone will stop their music subscription. They found that for some users who manually renew, having data from about 45 to 90 days improves prediction but adding more days gives less extra benefit. However, when they changed the way they looked at the problem, the best number of days to use also changed, meaning there is no one-size-fits-all answer. They suggest that predictions depend a lot on how you define the task and what data you use. Their results come from one music streaming dataset and might differ in other cases.
churn predictionearly behaviorsubscriptioncohort designfeature setmanual renewaldiminishing returnsmusic streamingprediction windowdata sufficiency
Authors
Xiao Han, Yao Xiao, Chenyu Wu, Tongchen Zhang
Abstract
How many days of early behavior suffice for subscription churn prediction? In the public KKBox dataset, the early indicator of churn is typically an indicator of someone's contract status; however, when looking in the heavily churned manual-renewal segment, having access to early behavior creates a substantial increase in prediction for that specific segment (PR +0.10 at 120 days). A nine-window sufficiency curve shows a diminishing-returns knee in a 45-90 day band. However, stress-testing over three cohort/task designs shows that this curve is singular to the design being tested; for example, in our test with a moving target, the curve inverts and can shift depending on the feature set used. Therefore, any window-sufficiency claim should state its cohort construction, target definition, and feature families. All evidence is from one music-streaming dataset; the mechanism should generalize but the magnitudes may not.