Look at a cohort chart for almost any supplement brand and there is a step in it. Month one is fine. Month two holds. Somewhere in month three the line drops, and it drops for subscribers who never opened a support ticket, never complained about the product, and never asked for a discount.
That shape is not a product failure. It is the gap between when a slow-result product starts working and when a customer decides it is not working — and in most brands, nothing at all is scheduled into that gap.
Why does month three specifically?
Because the promise and the proof are on different clocks.
Joint, gut, sleep and skin supplements are sold on a change the customer is supposed to notice: sleeping through the night, less bloating, a knee that tolerates a run again. Those changes are gradual, and the category itself sells them on a window of weeks rather than days — read the back of almost any bottle. The same clock governs anything whose result is slow to show: a retinol serum, a hair routine, a greens powder, a mobility programme.
A subscriber’s patience runs on a different clock entirely. It runs on the credit card statement. The first charge is optimism. The second is fine. By the third, the question has changed from “is this working” to “can I justify this”, and the honest answer for most people is that they cannot see anything yet.
So the cancellation is not a verdict on the product. It is a verdict on the absence of evidence, delivered at the exact moment the product was about to start producing some.
What is actually going wrong in the flows?
Three things, and they compound.
The education stops after the welcome series. Most brands in this category have a good week-one flow. Then nothing until the reorder notice. The subscriber is left alone for the entire period during which they are forming a judgement, and the only message they receive in that window is a charge notification.
Nobody is measuring adherence. A supplement only works if it gets taken, most days, at roughly the right dose. A bottle sold as a 30-day supply at two capsules a day only lasts 30 days if both capsules go in. Miss three days a week and it stretches past seven weeks — product piles up, and the subscription gets paused for a reason that reads as “too much product” but is actually “we never helped them build the habit”.
There is no way to record a small win. The subscriber who noticed in week six that they were finally sleeping through the night has nowhere to put that observation, so it never becomes part of the story they tell themselves at month three. Brands that ask — a single question, at the right moment — turn a vague impression into a stated one, and people are considerably less likely to cancel something they have just told you is working.
What should be running in that window instead?
Four things, all of them scheduled against the subscription start date rather than against a purchase event.
A day-structured education track, not a newsletter
Messages timed to the adherence and observation milestones of the specific product, not to a marketing calendar. Week one is about building the routine. Week two to four is about what normal looks like — including the honest version, which is usually “nothing yet, and that is expected”. Week five onwards is about what to watch for, described concretely enough that a customer can recognise it.
Setting the expectation early is the highest-value message in the whole track. Someone told in week one that they should not expect to see anything for several weeks does not treat week six as failure.
A dosing habit loop
Reminders that decay. Daily for the first fortnight, then weekly, then gone. Tie the reorder timing to actual consumption rather than to a fixed billing interval, which is where a properly modelled replenishment window pays for itself twice: fewer pauses caused by surplus, and fewer stockouts caused by a customer doubling the dose.
One progress check-in
A single question, sent at the point where a change should first be observable. Not a survey. One tap: better, the same, or worse.
“Better” gets acknowledged and stored on the profile, and it is the most useful save asset you will ever collect — you can show it back to the same person in the cancel flow.
“The same” gets routed into content that resets expectations and, where relevant, checks the dose. “Worse” gets a human. That last branch matters more than its volume suggests, because a worse answer on an ingestible can be a real problem and it should never be answered by an automation.
A cancel flow that knows which month it is
A subscriber cancelling in month three is a different case from one cancelling in month twelve, and the flow should treat them differently. Month three needs the expectation reset, an offer to reduce the dose or extend the interval before cancelling outright, and — if they gave you a “better” answer six weeks ago — that answer, quoted back.
What it should not be is a blanket discount. Discounting everyone at the point of cancellation trains subscribers to cancel, and it costs the most on exactly the cohort that was going to stay anyway. The cancel-flow logic is worth building properly rather than accepting whatever the subscription app ships with.
What about the subscribers who leave for a billing reason?
A meaningful share of what looks like month-three churn never involved a decision at all. Between 20% and 40% of subscription churn is involuntary (Paddle / ProfitWell) — cards that expired or were reissued, not customers who left.
Before rebuilding education flows, split the cohort. If a third of your month-three cancellations are failed payments, no amount of content fixes them. That is a separate build, and we wrote the whole thing up in the failed payment recovery guide. Do that first: it is faster, it is cheaper, and it makes the remaining churn number honest enough to work with.
Where this sits in the stack
Everything above lives in Klaviyo, triggered from your subscription platform’s events, with the check-in response written back to the profile so the cancel flow can read it. If those events are not arriving reliably — and in a Shopify plus Recharge plus Klaviyo stack they often are not — the flows will look broken when the data is what is broken. The order-sync diagnostic is the place to start on that.
More on how this plays out when you bill on a schedule on our subscription brands page, including what the stack usually looks like and what we build first.
Sources
- Paddle / ProfitWell — the finding that between 20% and 40% of subscription churn is involuntary. Direct article link to confirm before publication.
The category timing described here reflects how these products are labelled and sold, not a measured benchmark. Our own cohort figures for month-three churn on slow-result products are in collection and will be published, with method and sample size, on the benchmarks pages.