"We are acquiring more customers than we ever have and the subscriber base is flat. Do we spend on winning people back, or on acquiring more?"

It reads like an allocation question. It is not. It is a diagnosis question wearing an allocation costume, and it cannot be answered until you know where the churn was created.

Most brands only have 2 boxes to put churn in. That is the problem.

The standard split is preventable churn and product-driven churn.

Preventable covers billing confusion, product buildup, no flexibility, no onboarding, failed payments. Fixable with systems.

Product-driven covers the product not working, not tasting acceptable, or a competitor being better. Not fixable with marketing.

It is a clean model. It also leaves the largest category in consumable subscriptions with nowhere to sit.

Acquisition churn. The customer was acquired on a promise the product was never going to keep on that timeline, or was never the right customer in the first place.

Product churn means the product failed. Acquisition churn means the promise failed. Those are different problems with different owners, and almost every brand files the second as the first, because the customer says the product did not work and everybody takes them at their word.

That misfiling is expensive. It sends brands into reformulation, repackaging or discounting a product that was never the problem, while the thing that manufactured the churn keeps running in the ad account with budget behind it.

Cancellation reasons are the input everyone uses and the input nobody audits.

"Too expensive" almost never means the price. It means not worth it at this price, which is a value judgement. Discount in response and you convert a value problem into a margin problem while keeping the value problem.

"I did not feel anything" is a timeline statement, not an efficacy statement. On a compound with an 8 to 12 week mechanism, the customer is reporting the gap between the onset they were promised and the onset they got.

And if your dropdown offers price, delivery and did not work, every reason on earth collapses into those 3. The menu is writing your data.

Read the stated reason against 2 behavioural facts instead: how many orders they took before cancelling, and whether they engaged with onboarding at all. A cancel at order 2 with zero onboarding engagement is not product churn, whatever the free text says.

Active subscribers is not the same as billed subscribers. Skips, pauses, pushed billing dates and cards sitting in dunning all live inside the same active flag in most subscription apps. A brand can hold a flat subscriber count while paid orders fall for 2 quarters.

Pull 4 numbers separately every month: billed successfully, skipped or paused, in dunning, cancelled.

Failed payment recovery is the cheapest retention work in the business and it is routinely logged as churn. Fix dunning before you run a single win-back campaign. It is not marketing, it is collection, and the yield is higher.

This is the one that matters if you are pushing spend right now.

Take month 1 and month 3 retention for your last 3 monthly cohorts. Put them against the same 2 points for cohorts from 6 and 12 months ago. Read the direction, not the absolute number.

Cohort decay under scale is the clearest signal of acquisition churn there is, because the product did not change and the retention system did not change. Only the customer and the promise did.

You do not grow because you acquire a lot of customers. You grow because acquired customers stack. Retention is what makes speed safe. Without it, speed kills, because you burn through your addressable universe faster than you can compound value out of it.

It is set in the ad and the PDP, and it detonates on the billing date.

Nir Eyal's Experience Loop, set out in Beyond Belief, runs in 4 steps: believe, anticipate, feel, confirm. A belief sets the frame. The body prepares for what it expects. The experience is then read through that frame rather than measured against nothing. Then the customer confirms the story out loud, to themselves first.

The evidence is not soft. In a Caltech and Stanford study published in PNAS, identical wine was rated as more pleasant, and showed higher activity in the brain region that encodes pleasantness, when subjects were told it was expensive. A BMJ study on branded analgesics found branded tablets relieved headaches significantly better than identical unbranded ones, with the effect strongest in the group given the placebo. Eyal's book catalogues more of the same, including decaf drinkers told they had caffeine reporting greater alertness.

The commercial reading is simple. That loop runs whether you design it or not. Every acquisition claim installs an expectation. Every billing date schedules a verdict. Most brands have pointed the 2 straight at each other, then wonder why month 2 is where the base falls out.

A 7-day promise on a mechanism that takes 8 weeks does not produce a disappointed customer. It produces a scheduled one.

Here is the part that gets missed. The onset window is often not a marketing choice at all. It is a consequence of the delivery format, decided long before anyone writes a headline.

Creatine is the clearest worked example, because the science is settled and public. There are 2 routes to muscle saturation, both documented in the ISSN position stand. Load at roughly 0.3g per kilo of bodyweight a day, about 20 to 25g, for 5 to 7 days and you are saturated inside a week. Or take 3 to 5g a day with no loading phase and you arrive at the same place in about 28 days.

Same destination. One route takes a week, the other takes a month.

Now put that compound in a gummy at 1.5g a unit. A loading protocol becomes 13 to 17 gummies a day for a week. Nobody prints that on a label and nobody follows it. The fast route is unavailable, so the clock is fixed at 28 days by the format, before a single decision is made about pack counts, price points or retail listings.

The principle generalises well past creatine: The format sets the clock. The clock sets the promise you are allowed to make. And the pack has to outlast the clock.

A pack that holds fewer days of product than the mechanism needs is not a small pack. It is a scheduled disappointment with a barcode on it. The customer takes it exactly as instructed, for as long as it lasts, observes nothing, and reaches a conclusion that is scientifically wrong and behaviourally permanent. They do not repurchase, and they tell people, because a debunking travels further than a mild success.

That is acquisition churn manufactured at the shelf rather than in the ad account. It is the same category and it has the same owner: whoever decides what goes in the box.

Two consequences worth sitting with. First, in a category where the mechanism is slow, adherence is the product, so it is the thing to measure and reward. Loyalty schemes that reward money spent rather than days taken are pointed at the wrong behaviour. Second, if you are tempted by a money-back guarantee to prove confidence, price it before you announce it. It is a genuine forcing function, because you cannot honour it without building the tracking, and the tracking hands you a number nobody else in your category has. But on a sub £30 unit, refund handling plus the lost stock often exceeds the gross margin, and most slow-mechanism compounds have a documented non-responder population who will claim in good faith.

This is not an argument for manufacturing belief you cannot honour. Setting an expectation the product cannot meet is what created the problem in the first place. The work is to promise a window the mechanism can actually keep, then give the customer something real and early to notice while the primary benefit is still being built.

One tell worth checking today. If your review request fires on a delivery webhook, you are asking on day 3. You are asking a customer to confirm something they have not experienced yet. They either write nothing, which costs you the review, or they write something lukewarm, which hardens a soft opinion into a stated position they now have to defend to themselves. Move it to day 21 or later on slow compounds.

The usual comparison is cost to win one back against cost to acquire one new. That comparison is wrong and it produces confident bad decisions.

The right comparison is forward margin against forward margin. A reactivated subscriber has already churned once, and second-life cohorts retain worse than first-life cohorts. A win-back at half the CAC can still be the worse purchase once you model 12 months out.

Reactivate only where all 3 hold: the churn was system churn, the system failure is already fixed, and the lapse is inside roughly 90 days. Past that, a lapsed subscriber is a cold lead carrying a bad memory, and you have to overcome the memory before you can make the sale.

Never win them back with a deeper discount than the original offer. It confirms they were right to leave, teaches your base that cancelling is a pricing lever, and returns them on worse unit economics than the ones you lost.

Acquisition is the easy half. These operators run the hard part on TikTok Shop.

The partners I work with run TikTok Shop for brands you already know. Grüns, Goli and Vita Coco in the US. Pukka, Proper and ManiLife in the UK. Both are official TikTok Shop agency partners, one per market, and I only make introductions I would make for myself. Tell me where you are in two minutes and I will match you to the right one.

Get matched to a TikTok Shop partner →

1. Split your churn 3 ways, in percentages. Not 2. Pull last quarter's cancellations, cross-reference every stated reason against order count and onboarding engagement, and put a number on each category. You cannot allocate against a split you have never seen.

2. Run the cohort direction check before you raise spend again. Last 3 cohorts against cohorts from 6 and 12 months ago, at month 1 and month 3. If recent cohorts retain worse, more spend fills a leakier bucket faster.

3. Put 3 numbers next to each other, per SKU. Days of product in the pack, at the dose you actually recommend rather than the minimum one. Days until the mechanism delivers something the customer can notice. And the timeline your top-spending ad implies. Most brands have never seen these 3 on the same line. Any SKU where the pack or the ad is shorter than the mechanism is a SKU that is producing its own sceptics.

Fix the biggest category first, not the easiest one. For most brands scaling right now, the biggest category is the one they do not have a box for.

If this was useful, forward it to a founder who is about to double their spend.

I do 3 free cohort analyses a month.

You give me read only access to Shopify Analytics and Reports. I go through your cohort data, your pack economics and your repeat curve, and come back inside 2 working days with what I find. Then a 30 minute walkthrough of it.

No charge, no obligation, nothing to sign.

It is built for brands doing 6 figures a month with a subscription programme already running

Or apply here → apply.consciouscommerceco.com/diagnose