How Sociaaal grows

Dozens of A/B tests a month per app (~100/month across the portfolio). Focus early on the top of the funnel (first screens, adding/removing onboarding screens), then test all around the app once an app has ~1M users/month. Each test starts at small scale and is expanded if promising. Of ~20 tests: 10 neutral, 5 bad, 3 promising, 2 winners.

Look for apps that look like short-term viral hits but tap a broad, lasting interest (e.g. celebrity lookalike curiosity). Buy them after the viral peak while revenue is falling and sellers don't know how to grow them. Judge on data: can we scale it profitably long-term and 20x revenue, making multiples of the acquisition price. Then run the studio's UA machine plus dozens of A/B tests a month per app to lift ARPU (the LTV/CAC loop).

All ads fully AI-made for 2+ years, including screen-recording styles: ~4,000 video ads, hundreds of playable ads and thousands of statics per month. 6-person ads team mostly generates new ideas and prepares material; agentic video editing via Poolday once an angle is nailed; human video editors for one-offs, new concepts and taste; others build playable mini-game ads. Scaling volume required more diverse ideas, so the creative team grew 3x in 6 months.

Don't treat a losing test as permanently settled: results depend on the surrounding flow, which keeps changing. Periodically retest old assumptions, especially high-lift ones, in the current configuration. The Celebs paywall-removal test failed several times before winning.

Treat creative as an infinite search space: spread tests across many new angles, concepts and formats; only when one works, make variations and go deeper there. Don't copy the market leader (they are the leader in most niches). Brute-forcing is impossible, so combine taste, data and process.

Combine a top-down market view (structure, willingness to pay, growth) with a bottom-up test: run ads for apps that don't exist yet or that are acquisition targets. Ads are highly standardized so results compare against a large internal benchmark. Then check the target's own ARPU, retention and marketing data.

Every A/B test gets a stated reason (hypothesis), result and learning in a backlog, never 'it feels better'. Learnings are generalized into strategies (e.g. 'remove onboarding friction when down-funnel conversion is strong') with the 5-7 concrete changes that achieved it, so an agent or new hire can find the equivalent in other apps.

AI surfaces videos (TikToks etc.) from completely different industries; creative analysts review them and, when relevant, turn them into ad ideas for the studio's apps, rather than copying the niche leader.