Subscription churn is decided early. The customer who cancels in month six usually made the decision in week five, when the second delivery arrived before they had finished the first, or arrived at a moment they could not afford, or arrived without them remembering they had signed up.
That is why retention programmes aimed at long-tenured subscribers so often disappoint. They are addressing a population that has already survived the dangerous part. The leak is at the start, and it has three causes: the timing is wrong for how the product is actually used, the subscription is harder to change than it is to cancel, and nothing happened between the first and second delivery to make the arrangement feel like a relationship.
So we treat the first three cycles as a single design problem, separate from everything that comes later. Get those right and the retention curve after them flattens on its own.
Set the interval from consumption, not from convenience
The default interval on most subscriptions is chosen because it is tidy, not because it matches how quickly the product runs out. That mismatch is the largest single cause of early cancellation, and it is entirely preventable.
Work out how long a unit actually lasts in normal use, then look at your own reorder data from one-time buyers to check. If people who buy manually come back on average every seven weeks, a four-week subscription will bury them and an eight-week one will leave them empty. Those are different failures and they cancel for different reasons.
Where usage varies between customers, ask at signup. A single question about how much they use gets you a better starting interval than any assumption, and asking signals that you intend the arrangement to suit them.
Then watch for the signal of a wrong interval, which is a skip. A customer who skips is telling you they have too much. Treat that as information and offer to change the schedule permanently rather than just deferring the same problem by a month.
The customer who cancels in month six usually decided in week five.
Make changing easier than leaving
Most cancellations are not rejections of the product. They are the only control the customer could find.
So the account area has to make the alternatives at least as easy to reach as the cancel button. Skip the next delivery. Delay it by two weeks. Change the quantity. Swap the flavour or size. Pause for a month. Change the payment date to just after payday.
When those options are buried and cancel is prominent, you are converting a scheduling problem into a lost customer. When they are visible, a large share of the people who came to cancel do something smaller instead.
The cancellation flow itself should ask one question about why, with the alternatives offered in response to the answer given. Too expensive gets a smaller size or a longer interval. Too much product gets a skip or a change of schedule. Not using it gets help with using it. Only the customer who genuinely does not want the product should reach the final confirmation.
And let them cancel when they insist. A trapped customer disputes the charge, and a dispute costs more than the subscription was worth.
The gap between the first and second box
Nothing arrives in that gap for most subscribers, and it is the period when the decision is made.
What should arrive is help with the product. How to use it properly, how much to use, what to expect in the first weeks, what other customers do with it, what to do if something is not working. Every one of those messages reduces the chance of a cancellation that is really a disappointment about results.
Then a message before the second charge, sent early enough to be useful rather than as a notification after the fact. Tell them what is coming, when, and how to change it. Stores worry that this reminds people to cancel. In practice, an unexpected charge produces a cancellation and often a chargeback, while an expected one produces a skip at worst.
And the second delivery itself is worth designing. A small note, a sample, a suggestion of what to try next. The first box is exciting because it is new. The second is the one that has to feel like a decision worth keeping, and it usually arrives with nothing in it but the product.
Failed payments are churn nobody chose
A meaningful share of subscription losses are not decisions at all. A card expired, a bank declined a recurring charge, an address changed. The customer still wants the product and simply stops receiving it.
This is the cheapest retention work available, because there is nobody to persuade. It needs a retry schedule that spreads attempts over days rather than hammering the card, a message that tells the customer plainly what happened with a direct link to update the card, and a backup channel so a failure is not only reported by email.
Card updating services offered by payment processors handle a portion of expiries automatically and are worth enabling.
The number to watch is the share of cancellations that came from payment failure rather than from a request. If nobody has separated those in your reporting, they are hiding inside your churn rate and making your product look less loved than it is. We separate them in the first week of any subscription project, because the fix is quick and the recovered revenue funds the slower work.
Measure cohorts, not a monthly rate
A single churn percentage for the whole subscriber base is almost useless, because it blends people in their first month with people in their second year, and those groups behave nothing alike.
Measure by cohort. Take everyone who started in a given month and track what share is still active after one, two, three and six cycles. The shape of that curve is where the information is, and it lets you see whether a change you made actually improved retention for the people who joined after it.
Watch the drop between the first and second delivery as its own number, because that is the single most sensitive indicator in the whole programme and it responds quickly to the fixes described above.
Also separate voluntary from involuntary churn, and separate people who cancelled from people who paused, because a pause that returns is not a loss.
And judge the value of a subscriber on the number of deliveries they complete rather than on the sign-up count. A promotion that doubles sign-ups and halves the number who reach a third delivery has cost you money, and only cohort reporting will show it.