When retention falls, a discount is an easy response to propose. It is also an expensive way to solve the wrong problem. A customer who never understood the product, could not complete the first task or left because a payment failed may need something quite different.
First establish who is leaving, when they leave and what changed for that group. Then test a fix against the value of the additional customers it actually retains.
A retention framework should make those steps explicit. It should also stop you from treating a blended percentage as a diagnosis.
Define what counts as a retained customer
Suppose a meal-planning subscription doubles customer acquisition, but 12-month paid-customer retention falls from 88% to 78%. Management wants to launch renewal discounts.
Before assessing the proposal, confirm the metric. Are these customers still paying 12 months after joining? Have they renewed an annual contract? Does a paused subscription count? Are failed payments included as churn immediately or after a recovery period?
Keep paid retention separate from product usage. A person can pay for an annual plan and stop using it after a week. That person is financially retained for the period but may be unlikely to renew.
Also distinguish customer retention from revenue retention. Losing several small accounts while expanding a large one can produce weak customer retention and strong revenue retention. Choose the measure that fits the question, then examine the other if it changes the economics.
Compare customers at the same age
A cohort is a group whose experience you follow from a defined starting point, such as first payment. Compare its retention after one month with other cohorts after one month, and its 12-month retention with cohorts old enough to have reached 12 months.
Do not compare a new customer's second week with an established customer's second year. Nor should you record customers who have not yet reached a milestone as failures at that milestone.
A cohort table helps separate changes in the product from changes in the age of the customer base. Read across a row to follow one group over time; compare the same time-since-joining column across groups to see whether newer customers perform differently.
For the worked example below, assume both acquisition cohorts have completed a full 12-month observation period.
Check whether the customer mix changed
Suppose the old cohort contained 1,000 organically acquired customers, with 88% retained after 12 months. That leaves 880 retained customers.
The new cohort contains 2,000 customers: 1,000 organic and 1,000 acquired through affiliates. Organic retention remains 88%, while affiliate retention is 68%.
| New-cohort channel | Starting customers | 12-month retention | Retained customers |
|---|---|---|---|
| Organic | 1,000 | 88% | 880 |
| Affiliate | 1,000 | 68% | 680 |
| Total | 2,000 | 78% | 1,560 |
The blended rate fell because the business added a lower-retention channel. Organic performance did not deteriorate.
The company also has more retained customers in absolute terms: 1,560 compared with 880. Neither observation tells you whether the affiliate channel is attractive. You still need its acquisition cost, customer contribution and potential for improvement.
This changes the next question. Investigate the affiliate proposition and customer experience rather than discounting the entire base because the average fell ten percentage points.
Quick Exhibit Drill
A pet insurer's 12-month renewal rate has fallen, and the board wants to cut its spending on price-comparison sites.
- Cells show
- New customers at 12-month renewal rate
| Year joined | Direct | Comparison sites | All customers |
|---|---|---|---|
| 2024 | 3,000 at 85% | 1,000 at 61% | 4,000 at 79% |
| 2025 | 3,000 at 85% | 3,000 at 63% | 6,000 at 74% |
What best explains the fall in renewal across all customers?
Follow the journey to the first useful result
Suppose many affiliate customers pay but never create a first weekly meal plan. Look at what the affiliate promised, what those customers expected and where the first-use process breaks down.
Perhaps the promotion attracts people seeking a one-off recipe download. Perhaps setup requires information they do not have. Perhaps the plan ignores dietary preferences they assumed it would handle.
Those are different problems. Review acquisition messages, customer interviews, support contacts and the steps completed before departure. Use the evidence to choose a specific change: clearer qualification, simpler setup or a better first result.
A pattern in usage data is a useful lead. It does not establish that forcing every customer through the associated action will improve retention. More motivated customers may both complete setup and stay longer. Test whether helping comparable new customers reach the first useful result changes their later behaviour.
For the meal-planning business, a guided first-plan experiment is a more direct response to this finding than a discount arriving at renewal months later.
Separate voluntary departure from payment failure
Some customers choose to leave because the service no longer meets their needs. Others lose access after an expired card, a temporary payment failure or a billing problem.
These should not automatically enter the same intervention. The first group needs a better value proposition or experience. The second may need a clear way to update payment details or resolve the failed transaction.
Check what the retention metric records during recovery so you do not count the same customer as a loss and a new acquisition when payment resumes. Understanding the event definition can be as important as analysing the percentage.
Count the renewals an offer actually creates
Consider a separate renewal test. Without an offer, 700 of 1,000 eligible customers are expected to renew. Each renewing customer contributes £40 over the next period before any retention discount.
A £10 renewal discount increases renewals to 760. Assume every renewing customer in the offer group receives the discount, and all other contribution and cost assumptions remain the same.
Without the offer, contribution is 700 × £40 = £28,000. With the offer, it is 760 × £30 = £22,800. Retention improves, but contribution falls by £5,200.
The offer creates 60 additional renewals worth £2,400 before discounts. It gives away £7,600 across all 760 renewals, including those that would have happened anyway.
This is why “we saved 760 customers” would be the wrong conclusion. A credible comparison group estimates the renewals that would occur without the offer. The incremental benefit comes from the difference.
A smaller, better-targeted offer could work. So could a product improvement with a lower cost per additional retained customer. Compare them on the same period and include their implementation costs.
Quick Case Math Drill
A climbing-centre chain will give a £10 credit to every annual-pass holder who renews, and is testing it on 1,000 expiring passes.
- Renewals without credit
- 800 expected
- Renewals with credit
- 900 expected
- Contribution per renewal
- £120 before the credit
What net extra contribution does the credit create across the test group?
Choose the fix and the evidence that would justify expanding it
For the meal-planning example, first validate the affiliate mix and investigate the failure to create an initial plan. Test the specific acquisition or onboarding change that addresses the observed barrier.
Measure the immediate behaviour it should change, such as completing a useful first plan, then follow paid retention and contribution as the cohort matures. Early activation is a leading indicator; it is not a substitute for observing later renewals.
Keep the decision tied to economics. A channel with lower retention may still create valuable customers at low acquisition cost. A high-retention channel may be unattractive if customers are expensive to acquire and serve. Use the growth framework to connect the result to the broader plan.
Try Streaming Content Investment to practise the reasoning in another business. Explain which subscribers the investment might retain, how you would estimate additional retention and what contribution is needed to justify the spending.
Structure drill
Would your case structure convince an MBB interviewer?
Streaming Content Investment
Test whether a content investment can reduce churn enough to pay for itself. Present your approach to Casey, Case Room’s AI interviewer, then review feedback on your structure and starting point.
Frequently asked questions
Defining and measuring churn correctly
What should I clarify first when a case says customer retention has fallen?
Clarify what retained means before analysing causes. In the meal-planning example, ask whether customers must still be paying 12 months after joining and whether an annual renewal counts. Also ask how paused subscriptions are treated and whether failed payments count as churn immediately or after a recovery period. A change in definition can look like a change in behaviour.
Do paused subscriptions and failed payments count as churn?
It depends on the client's definition, so ask rather than assume. A paused customer is not paying but may return. A failed payment may be recovered within a grace period. Keep paid retention as its own measure so you know whether the decline reflects lost revenue or different counting rules.
What is a cohort, and why compare customers at the same age?
A cohort is a group you follow from a defined starting point, such as first payment. Compare one-month retention with other cohorts at one month. Compare 12-month retention only with cohorts old enough to have reached 12 months. Comparing a new customer's second week with an established customer's second year confuses early drop-off with long-term loyalty.
How should I treat recent customers who have not reached 12 months yet?
Leave them out of the 12-month figure rather than counting them as retained or lost. They have not yet had the chance to reach that point, so including them distorts the result. You can still compare their early retention, such as at one month, with older cohorts measured at the same age.
Finding out why customers leave
Can overall retention fall even if no customer group gets worse?
Yes, if the customer mix shifts. In the example, organic customers retain at 88% in both cohorts. The new cohort adds 1,000 affiliate customers who retain at 68%, giving 1,560 retained out of 2,000, or 78%. The headline rate falls, but organic behaviour is unchanged. The business also retains more customers in total than the previous 880.
Does a lower affiliate retention rate mean the channel should be cut?
Not on retention alone. Affiliate customers lower the average rate, but they still add 680 retained customers. Whether the channel is worthwhile depends on what those customers cost to acquire and what they contribute, and the example does not provide those figures. The more useful question is why affiliate customers leave more often and whether that gap can be closed.
If customers pay but never use the product, where is the problem?
Follow the journey to the first useful result, such as creating a first weekly meal plan. The promotion may attract people who want a one-off recipe download. Setup may require information they lack, or the plan may ignore their dietary preferences. Each is a different problem, so review what acquisition promised alongside the onboarding process.
Should customers whose card failed receive the same retention offer?
No. Involuntary departures need a different fix. Someone who chose to leave needs a better value proposition or experience. Someone who lost access through an expired card or a billing problem may only need an easy way to update payment details or resolve the failed transaction. A discount for them spends money without addressing the cause.
Deciding whether a retention offer pays
How do I test whether a renewal discount pays for itself?
Compare total contribution with and without the offer, not just the renewal rate. The discount usually goes to every renewing customer, including those who would have renewed anyway. The extra contribution from additional renewals must therefore exceed the discount given to the whole renewing group, with all other cost assumptions held constant.
Does the £10 discount work in the article's example?
No. Without the offer, 700 renewals at £40 give £28,000. With the offer, 760 renewals at £30 give £22,800, so contribution falls by £5,200. The 60 extra renewals add value. However, giving £10 to the 700 customers who would have renewed anyway costs more than that extra value.
How many renewals would the discount need to break even?
Under the same assumptions, contribution with the offer must reach £28,000 at £30 per renewing customer. That requires roughly 934 renewals out of 1,000 eligible customers, far above the 760 the test produced. Stating the threshold this way shows the interviewer how far the offer is from paying for itself.
What should I recommend in the meal-planning case?
Hold the renewal discount and start with the diagnosis. Check that the affiliate mix explains the fall, then investigate why affiliate customers fail to create an initial plan. Test the specific acquisition or onboarding change that addresses the barrier you find. This targets the cause instead of paying every renewing customer to stay.
Is better activation enough evidence to roll out a fix?
No. Activation is a leading indicator, not a substitute for retention. First measure the immediate behaviour the fix should change, such as completing a useful first plan. Then track paid retention and contribution as the cohort matures. Expand the fix only if those later outcomes improve, not merely because early engagement looks better.