How PropGPT grows

Used PostHog analytics, Superwall paywall data and user conversations to find behavior that didn't match the app's design. High install-to-trial (45%) with low trial-to-paid (13%) meant users wanted the promise but the product experience disappointed. Also: track onboarding drop-off to find screens that don't sell, and track feature clicks to find the most compelling value proposition to use in marketing.

The first version made users go through sportsbooks, input their own bets and check whether each was good. Realized users just wanted to be given the right picks ('the answers to the test'). Shut down all marketing and spent four months on engineering and design only, rebuilding from scratch around pre-analyzed daily picks from their ML models. Relaunched April 15.

At first launch (mid NFL season), drawing on experience with other successful app founders, spent a few thousand dollars on influencer marketing to measure how many downloads the money bought.

About a month after the April 15 relaunch, went all in on marketing again timed to the NBA Playoffs; marketing now runs about $10K/month.

Scaled with influencer marketing because followers engage deeply and convert when someone they admire uses the product; kept publishing videos (around 70 by the time one broke out) for the chance of a viral hit.