The full arithmetic, and the pilot that tests it
Three chapters, six minutes. By the end you will know how far the market has already moved, what the model returns end to end on the reference book, and what a two-week slice of your own base has to prove before anything scales.
Recovered margin plus the one-time settlement release, at the 55% base case on the $5M reference book — the pilot replaces the base case with your measured rate.
Get the one-page summary, written for your role.
The validation
The app stays the product surface. The web becomes the commercial surface. That pattern is already settled at the top of the market.
41%
of the highest-revenue subscription apps already generate revenue on the web.
Source: RevenueCat, State of Subscription Apps 2026 — apps with $25M+ annual gross consumer spend.3
Spotify routes Premium signup and plan management through spotify.com. Netflix stopped accepting App Store billing for new and returning subscribers in 2018. None of them needed a policy change to do it.
One distinction worth being blunt about: a web-billing SDK opens a rail for new subscribers. It cannot move your existing book. Migrating a live base is an operational campaign — cohorts, holdouts, billing transitions, support — not an integration. The two are complementary, and only one of them is what this playbook is about.
Why not the DMA’s in-app doors
In the EU the store fee doesn’t disappear under alternative billing, it reshapes — core-technology and store-services fees stack to roughly 17% — eligibility differs by region, and every variant changes your binary. The out-of-app lane is the conservative superset: identical in every market, no binary change, no reshaped fee. Full policy analysis in the docs →
A worked example
A model you can check beats a claim you have to trust.
Permanent, from each migrated subscriber's next renewal onward.
The settlement queue drains from 48 days to your Stripe payout schedule.
At 55% of the reference book accepting. Figures carry the payment rails only and are gross of the cost of running the migration5 — the audit nets both against your numbers.
The reference book is the same one Act I sized: $5M store-billed ARR, a 22% blended store fee, a 50/50 plan mix at $10 a month and $100 a year, priced on the Stripe schedule for your billing country. Nothing here is a new model — it is the same arithmetic, run to its conclusion.
One honest caveat sits under the whole figure. The migration rate is a prior, not a promise6 — 55% is this document’s base case, and it is the one number in this playbook that no amount of modelling can settle. That is precisely what the next chapter is for: the pilot exists to replace it with yours, measured on your own book against a matched holdout.
Proof before scale
Models persuade. A slice of your own book proves.
Migrated cohort
Offered the move. Cut on tenure, plan tier and engagement.
Matched holdout
Same tenure, tier and engagement. No offer. Left entirely alone.
the delta between them is the only honest number
Migration rate
What share of a reachable, engaged cohort actually accepts — measured, not modelled.
Churn delta
Cancellation in the migrated cohort minus the holdout, in basis points. This is induced attrition, isolated.
Net margin recovered
Annualised from the real accepted volume, on the real fee schedule.
However strong the worked example reads, it is still an argument, and the right answer to an argument about your revenue is an experiment on your revenue. Run it on a slice, measure it against a holdout, and decide on the result.
One thing does need planning: migration re-bases your LTV curves, so comparisons across the switch get awkward for a quarter. The new curve is the truer one — contribution measured on rails where 15–30% less leaks out — but the fix is structural, not cosmetic. New analysis axes for billing rail, migration cohort and wave, and migration events fed to your analytics stack as lifecycle events, so a migrated subscriber never books as a mysterious new acquisition.
The same shape repeats at every scale. A cohort’s first wave is a small proof before its own scale-up, which is what the wave engine in the next act is built around.
You have the size of the prize and the shape of the proof. Act III is the judgement and the machinery: the three decisions that move the rate, the cut that implements them, the gates that pace them, and what has to exist on the day the first wave sends.
3 RevenueCat, State of Subscription Apps 2026 — apps with $25M+ annual gross consumer spend.
5 These figures are gross of cost, in two ways. Running billing — dunning, tax, receipts, refunds, chargebacks — moves with the billing and costs something wherever it sits. So does the migration itself: vendor fees, internal engineering and support time, campaign cost. Neither is netted here. The audit nets both against your own numbers.
6 55% is this document’s base case — a modelling prior, not a promise. The pilot replaces it with your measured rate, on your own book, against a matched holdout.