Subscription Commerce Models That Survive Their Own Churn Maths

Subscription revenue is the most flattering line in commerce — recurring, forecastable, multiple-expanding — right up until churn math turns it into a leaky-bucket treadmill. Whether a subscription compounds or collapses is mostly decided by model choice and a handful of operational disciplines, all visible in advance.

The three models have different physics

Replenishment (coffee, razors, supplements, pet food) is the strongest physics: the need recurs on its own schedule. The work is matching your cadence to their consumption — wrong-cadence delivery is the top stated cancel reason in the category. Flexible frequency, easy skip, and consumption-informed defaults ("most two-cat households choose 6 weeks") beat discounts at retaining.

Access (memberships, content, service tiers) survives on ongoing utility; the danger is launch-value decay. The audit question: what did a member use this month? If the honest answer is "the discount", you run a coupon program with worse accounting.

Curation (boxes) has the hardest physics — novelty decays, per-box product cost is real, and churn cliffs at months 3–6 are structural. Survivors convert curation into discovery for a broader shop: the box recruits and educates; the catalogue retains. Pure curation with no second act is a cohort countdown.

Involuntary churn: the free retention program

A quarter to a third of subscription churn is failed payments, not decisions. The recovery stack — smart retries timed to salary patterns, card-updater services, dunning that treats the customer as a customer rather than a debtor, backup payment methods — reliably recovers a large slice at near-zero marginal cost. It is the highest-ROI work in the category and the most commonly unbuilt, because it lives between billing and engineering where nobody's OKRs point. Build it first; it outperforms every winback campaign you will ever write.

Pause beats cancel

Subscriptions with prominent, friction-free pause and cadence-change options retain materially better than those that hide cancellation — because "too much product" and "money is tight this month" are temporary states, and pause keeps the relationship alive through them. Hiding the cancel button, besides being an FTC "click-to-cancel"-era liability in the US and a consumer-law problem in the EU, converts temporary states into permanent exits with a grudge attached. Make leaving easy and coming back easier; the cohort curves reward it.

The metrics that predict the future

Monthly logo churn hides everything. Watch instead: cohort retention curves (does the curve flatten? at what month? the flat part is your business), net revenue retention including upgrades and cadence changes, payback vs the flat point (if CAC pays back after the curve flattens, you have a business; before, a bet), and cancel-reason distribution trended over time — it is the returns-reasons of recurring revenue, and it tells you which fix to build next.

Subscription mechanics — billing, retries, pauses, cadence logic, the data model under the curves — are software, and brittle subscription software shows up directly as churn. That layer is what we build; the physics above decide whether it is worth building on.