Subscription Paywall Testing on 60-Day Net Revenue

The number this engagement moves is sixty-day realized revenue per paywall view, net of refunds and failed billing retries. Trial starts are the metric most teams optimize and the one most likely to lie: a three-day trial behind a hard paywall in onboarding beats a seven-day soft paywall on trial starts nearly every time, and loses to it on cash two months later, once refunds and the first renewal attempt have both landed. A test started in March therefore reports in June, and the rhythm of this work is four or five real decisions a year rather than four a month. What gets measured: Paywall view to trial start, trial to paid, first renewal, refund rate by store and by country, and involuntary churn, meaning the share of cancellations that were a card failing rather than a person deciding. Involuntary churn is where the cheap wins are and almost nobody looks: grace periods left at a default, Play retry windows set to the shortest option, dunning messages that arrive in the wrong language. I read that before looking at a single paywall screen. Where the tests come from: Your cancellation survey text and your support inbox, read line by line, rather than a swipe file of other companies' paywalls. The list that comes out of a first pass is usually four or five items long, and they run in sequence rather than in parallel, because a paywall test and a pricing test running at once produce a result nobody can act on. Pricing: Price points get set explicitly per market rather than left to the store's automatic conversion, which turns a 39.99 anchor into an unrounded number that reads as an accident across half of Europe. The annual to monthly ratio gets a defensible reason behind it. Introductory and win-back offers are configured as store-native offers so they survive a reinstall, not as a code pasted onto a web page. In Quebec, the auto-renewal disclosure has to meet provincial consumer protection rules and the French has to be at least as prominent as the English. I check that, because store review sometimes does not, and a complaint costs more than the fix. What I will not test: Countdown timers on offers that do not expire. Decline buttons rendered in low-contrast grey. Pre-checked upsells. Any flow where cancelling takes more steps than subscribing did. These raise trial starts, and they raise refunds and one-star reviews alongside them. I would rather lose the engagement than build one. Wrong buyer: Apps under roughly three thousand paywall views a week per arm, where detecting anything smaller than a twenty percent swing takes a full quarter. For those the answer is qualitative work and a pricing decision made on judgement, which is cheaper, and I will say so rather than sell a test that cannot conclude. Teams who want a redesigned paywall screen delivered: the screen is the last five percent of this and a designer you already employ can draw it once the decisions exist. And anyone whose retention problem sits in the product, where no arrangement of a paywall holds users the app is losing in week one. Cadence and handover: Set-up and the first read take eight weeks. After that the rhythm is one test live at a time with a monthly written call on what to keep. Everything runs in your own RevenueCat or Adapty account against your store credentials. The handover document lists every test run, the sample it reached, the decision taken, and the tests I judged not worth a queue slot. That last list saves your successor a year.