Customer retention
Using churn models to improve pricing and customer retention
Turn churn predictions into better renewal, discount and customer-management decisions, and measure the contribution that retention actions create.
A churn model estimates the probability that a customer will leave within a defined period. That prediction becomes useful when it changes a decision: whom to contact, which renewal to review or where to investigate a service problem. For pricing teams, its value depends on connecting retention to contribution margin and measuring the effect of any intervention.
Key takeaway
High churn risk does not automatically justify a discount. Prioritise actions by their expected incremental contribution.
Define what leaving means
A cancelled subscription creates an observable churn event. In a repeat-purchase business, a long gap between orders may reflect seasonality, a lost account or a normal buying pattern. Set a definition and prediction horizon that match the action you can take.
A renewal team may need a view several weeks before a contract expires. A strategic revenue forecast needs a longer horizon. Use performance measures that reflect these different purposes, and check that the scores are calibrated for the population receiving them.
Six decisions a churn model can support
The strongest use cases connect a prediction to an owner and a specific next action.
- Renewal prioritisation: give account managers time to review important contracts before they expire.
- Retention offers: test whether an offer changes behaviour enough to cover its cost.
- Price-change reviews: identify segments that need closer monitoring after a price adjustment.
- Customer lifetime value: combine expected retention with future contribution and cost to serve.
- Service improvement: investigate recurring support or delivery problems in higher-risk groups.
- Portfolio planning: compare revenue and capacity scenarios under different retention assumptions.
Separate prediction from the effect of a price change
A model trained on historical data can predict churn well without telling you what will happen if you change a price. For example, customers receiving discounts may already have been selected because they were difficult to retain. That history can confound the relationship between price and churn.
Use controlled tests or an appropriate causal design to estimate the effect of a pricing intervention. Reserve a comparison group where feasible, record who was eligible and assess results over the period in which the commercial effect is expected.
An illustrative retention-offer calculation
Suppose a customer would contribute £1,000 over the next period if retained. A £100 offer is expected to increase retention from 60% to 70%. If the offer costs £100 for every retained customer who receives it, expected contribution changes from £600 to 70% × (£1,000 − £100) = £630. The expected gain is £30 before contact and delivery costs.
These are illustrative assumptions, not observed results. They show why the offer cost, the incremental retention effect and the payment conditions all matter. Discounting everyone with a high churn score can be expensive when many would have stayed anyway.
Start with one measurable workflow
Choose one renewal cohort, one intervention and one accountable commercial owner. Agree how contribution and retention will be measured, then compare outcomes with the baseline. Review whether the action reaches customers early enough to help.
Expand when the workflow demonstrates value and the team can operate it consistently. The model should support a customer-management process that learns from actual outcomes.
This Pricing Precision edition is adapted from 16 ways to generate value from churn models by Hans Christian Ekne at EKNE Consulting, originally published on 28 November 2024. It develops the practical implications for pricing and commercial teams.