How RevenueCat grows

A daily founder update email lists the 5-10 most important metrics (for apps: total paywall impressions). Each metric is charted as the daily value plus 30-day and 90-day rolling averages: 30-day above 90-day = trending up; daily values catch one-off spikes. No dedicated data analyst needed.

During YC, Jacob emailed fellow YC founders asking to put the RevenueCat SDK into their apps himself, met at a Blue Bottle for a few hours, sent pull requests that integrated the SDK and also fixed some of their UI bugs as an appreciation gift. Goal: make the first couple of customers truly happy.

Built from on-device receipt data across 22,000+ apps. Waited years until the customer base was representative of the App Store and the analysis was rigorous (data scientist plus business analyst). Everything averaged above a minimum threshold, apps with tiny data sets excluded, each app weighted equally, no scatterplots, and developers can opt out. Reported as quartiles plus median, not means. Released as a PDF at revenuecat.com/report, with a companion podcast, blog posts and a yearly update planned; readers asked to email feedback.

RevenueCat's in-app share-a-chart features were almost never used. Building a mobile app with an aesthetic, easy-to-screenshot home screen made customers post their numbers on X. Not designed as a sharing feature.

RevenueCat asks sign-ups 'How did you hear about us?' with options plus free-form answers, and the CEO reads every answer daily. Suggested for apps without onboarding: show it to ~5% of users with three big buttons and an easy dismiss.

Started with off-the-shelf Amplitude plus a few semantic events (app created, project created) and a generic page-loaded event with the path broken down, which allowed retrofitting later questions. When a restructuring of the product changed what core terms meant, added an ETL step that encodes each definition change by date into a normalized 'super definition'. Now close to running its own event tracking on top of its data warehouse.

RevenueCat was starting to build cost-control measures for its AI features (running on Anthropic's Opus); Jacob killed that work and chose to let usage run, betting inference prices will fall and using venture capital to absorb the cost meanwhile. Reference cost: one LLM call per app per day would be tens of thousands of dollars a month.

Architected from the start so costs scale with paying end users; non-paying small developers use the product free.

Posted on Reddit, then launched a beta, then framed each fundraise as 'RevenueCat's launching out of...' to put a narrative on ongoing building. If a launch goes poorly, launch again: few saw the first one.