Churn means different things in different businesses. Subscription cancellation is an observed event. A retailer usually sees only missing purchases and defines an inactivity proxy. These targets need separate names, horizons and evaluation.
| Contract item | Subscription example | Retail example |
|---|---|---|
| Decision unit | One active account and subscription episode at a weekly scoring time | One eligible customer at a monthly scoring time |
| Eligibility | Service active at as_of; account has not already entered the separate cancellation or win-back workflow | At least one qualifying purchase in the previous 90 days; enough observable history for the chosen cohort |
| Label | Effective service termination in (as_of, as_of + 30 days] | No qualifying purchase in (as_of, as_of + 60 days] |
| Feature window | Previous 7, 30 and 90 days, ending at as_of | Previous 30 and 90 days, ending at as_of |
| Label availability | Only after the horizon and the documented late-event allowance are complete | Same rule; a recent customer with no purchase yet is not a known positive |
| Action | Support investigation, payment recovery or an eligible retention offer | A re-engagement offer or no incremental contact |
| Business outcome | Incremental net contribution and retained accounts at a specified follow-up | Incremental net contribution and repeat purchases at a specified follow-up |
These are example definitions. Set inactivity windows from buying cadence, seasonality and the time available to act. A 60-day gap can be normal for an annual gift buyer. Record whether churn means voluntary cancellation, payment failure, non-renewal or any termination. Payment recovery often deserves its own model and workflow. Define grace periods, refunds, pauses, account merges and reactivation. When a cancelled subscriber returns, start a new episode.
Keep the target ahead of an actionable decision. A cancellation reason entered after departure is leakage. A cancellation notice submitted before scoring can be a valid signal for a clearly defined cancellation-rescue workflow, but it changes the prediction problem. Keep that workflow out of a model that claims to predict unannounced churn.
Write down the comparator too: existing rules, random allocation at the same capacity, or the usual retention policy. "Higher accuracy" is not a decision objective.