Skip to content

Stockertim Dictionary · Artificial intelligence · Business

Churn prediction

Also called churn modelling, churn model

In short

Estimating which current customers are likely to stop buying or cancel, usually with a model that learns from patterns seen before past cancellations.

  • Intermediate
  • Process
  • Checked

In detail

Churn prediction looks for customers who show early signs of leaving. For a subscription business, churn is a cancellation; for a shop, it may be a regular customer who stops buying. The aim is to spot these customers early enough for a retention message, a service call or a fix to make a difference.

A churn model is trained on past customers whose outcome is known. It learns which signals tended to come before a cancellation, such as fewer logins, a fall in order size, a complaint or a failed payment, and gives each current customer a probability of leaving within a stated period.

The scores are useful only if they lead to sensible action and are checked against what happens. A model that flags many customers who were never going to leave wastes offers, while one that misses the real leavers gives false comfort. Comparing the churn rate among flagged customers with the rate across all customers shows whether the model adds information.

Retention offers sent to everyone flagged can also train customers to threaten to leave. Many teams test retention actions with a holdout group, so they can tell whether the action, rather than the prediction, changed the outcome.

Worked example

Northwind Bikes has 5,000 subscribers. Its churn model flags 400 as likely to cancel next quarter.

  1. Overall churn last quarter: 300 / 5,000 = 6%
  2. Flagged subscribers who cancelled: 160 / 400 = 40%
  3. Cancellations caught by the flag: 160 / 300 = 53.3%
  4. Unflagged subscribers who cancelled: 140 / 4,600 = 3.04%

Flagged subscribers left at 40% against 6% overall, so the model adds real information, though it still missed 140 of the 300 who left.

In practice

Subscription, telecoms, banking and software firms use churn scores to plan retention calls, service fixes and win-back campaigns. Analysts track the churn rate by score band and test retention actions against a holdout group.

Watch out

  • A churn score is a probability for a group of similar customers, not a verdict on one person.
  • A drop in churn after an offer is not proof the offer worked without a comparison group.

Learn it properly

Official sources

Facts last checked on 3 November 2026.

Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.

English