Pause is a second exit

In July a kidney-health supplement and education brand, the one from the flat-revenue case, added a pause option to its subscriptions. The idea was the usual one: a person about to cancel gets offered a pause instead, keeps the subscription, and comes back. By the autumn 175 subscriptions had paused. Cancellations had not gone down.

The reason sat in the events. Of the 175 people who paused, only 2 had pressed cancel first (count, subscription events, July to October). The other 173 had gone straight to pause from the account page, where it sat next to skip and cancel as a third button. Pause had not intercepted anyone on the way out. It had become a quieter way to leave, with the subscription still counted as active.

What a pause is for

A pause is a save when it meets someone at the moment of the decision. Offered to everyone, all the time, it is a feature, and features get used by the people who want them, not the people you built them for. The person who would have kept paying now pauses for a month because the button was there. The person who was leaving still leaves, because nothing changed on the cancel path.

So the fix was not to remove pause. It was to move it to the cancel step, where it is offered once, to someone who has already said they are going, and to take it off the account page. That is the difference between a save and a second exit: a save is reached through the cancel button.

Where the save has to happen

The brand's cohort curve says when. About 70% of orders were on a subscription (share of orders, from order line items, June to July 2026). Following the subscribers forward from their first charge, about 18% cancelled in month one, 35% had cancelled by month three, and half were gone by month five. Tenure measured three ways disagreed, which is why I measured it three ways: a median of 4 to 5 months, a mean of 7 to 9, and about 12 if the current churn rate held.

Those months are the calendar for the save series. A winback that fires at day 60 for everyone meets nobody in particular. One that fires before the month-one, month-three and month-five decisions meets people at the moment they are deciding. The six subscription flows I wrote follow those months, with checkpoints at day 90 and month six, and a save series that offers pause or skip before any discount. They went live on 12 August 2026.

The order inside the save matters too. Pause first, skip second, discount last. A discount offered first teaches the whole base that cancelling is how you get one. Pause and skip cost nothing and answer the two commonest reasons people give: too much product, and money this month.

The flow that was missing people

While I was in the subscription events, the failed-payment flow turned out to be reaching 37 of 102 subscribers with a declined card (count, at the audit). Its check ran a moment before the store wrote the failed status, so for two-thirds of the people it existed for, it saw a healthy subscription and did nothing. A one-off series to the 58 it had missed brought 12 back. There was no control group, so I will not call that a lift, only a count.

The two problems are the same problem. The event the flow listens for and the moment the customer decides are not the same thing, and nobody had put them on one timeline. Pause fired when the button was pressed; cancellations happened on the cohort curve; the failed-payment flow fired a second before the failure existed.

Do this Monday

1. Export every pause and cancel event from the last 90 days with the subscriber, the date and the page it came from.
2. Count the paused subscribers who had started a cancel first. Divide by all who paused. At the brand it was 2 of 175, about 1%.
3. Count cancellations in the 90 days before pause launched and the 90 days after, as a share of active subscriptions. If it did not fall, pause is not saving anyone.
4. Pull your cohort curve: the share of subscribers still active at month one, three and five.
5. Put your save series send days on the same timeline. Mark each month where the curve drops and no message lands before it. Those are the gaps.
6. Move pause to the cancel step, offer it once, and take it off the account page.

In one line: a pause offered to everyone is a second exit; a pause offered at the cancel button, before the discount, on the months your cohort curve drops, is a save.