Pick Plug

Pick Plug sells one thing: a subscription that unlocks daily handicapper picks. When the subscription flow breaks, the business breaks — and it was breaking intermittently, which is the cruelest failure mode. Somewhere between 10 and 30 percent of purchase attempts were failing: a user would tap subscribe, the flow would error or the entitlement wouldn't arrive, and the revenue simply didn't happen. Intermittent means it passes a casual test, survives a cursory code review, and keeps bleeding conversions for months while everyone half-suspects the users are doing it wrong.

The backend had the mirror-image bug: the unsubscribe webhook handler was failing, so cancellations didn't reliably propagate — users who thought they'd cancelled were still being billed. Failed purchases lose revenue; failed cancellations lose trust and generate the one-star reviews and chargebacks that follow it. The app was leaking value at both ends of the subscription lifecycle.

Intermittent purchase failures are almost never one bug; they're a chain of small unsound assumptions that only align into failure under real-world timing. So the engagement started with an end-to-end audit of the whole lifecycle — the Android client's billing flow, the store's purchase states, and the server's validation and webhook handling — against production configuration, not a sandbox, because the original bugs had survived precisely by behaving in test environments.

The Android IAP validation flow was rewritten end-to-end to handle every state the store can actually produce, including the awkward ones — pending purchases, retries after interrupted connections, restored purchases on reinstalled devices. The unsubscribe webhook handler was fixed so cancellation events from the store reconcile server-side entitlements reliably. And since the daily-picks product lives or dies on delivery, push notification delivery was hardened in the same engagement — the subscription is only worth paying for if the picks actually arrive every morning.

In-app purchase conversion errors went to zero — from a 10–30% intermittent failure rate to none — and unsubscribe failures were eliminated. The client reported meaningfully improved retention and store reviews in the releases that followed, which is what you'd expect when the two worst moments in the user relationship (paying and leaving) both start working correctly. The engagement is small in the portfolio's word count and outsized in its shape: this is what mobile rescue work looks like — narrow scope, production-config diagnosis, and a fix measured directly in recovered revenue.