Roger Chen - How I Built the #1 App on The App Store (Twice)

Before the app was finished, co-founder Leo built a tappable fake of the app in ProtoPie (PowerPoint-like, not real code) and filmed it on a phone as a TikTok demo: a real texting scenario with Bro floating over the conversation. They posted it on the brand TikTok account and bought ads on it for fast reactions instead of waiting months for organic UGC. Roger frames ads as the tool for early feedback, not for scaling (paid gets pricier as you scale); a few thousand dollars is cheap next to weeks of building blind. Two outputs: at least one format that works, and a read on whether people want the product.

Bro's vision ('always with you') was too unspecific to market, so they looked for a concrete scenario. They tested 5-6 mock-up ad variations: some around the lock screen (e.g. Bro giving morning affirmations when you wake up in a bad mood), others around Bro floating over a texting conversation. They went all in on the floating-over-texts format, then dating apps, then miscellaneous life moments. The lock-screen versions also did well (they showed the 'proactive caring' idea could work) but used outdated designs, so they were dropped to avoid confusing viewers.

Classic hook and demo. The hook is a relatable scenario with a real person on screen (people stop for faces, swipe away without one); then the app interface. For Bro the demo reads like the iMessage thread itself, with Bro floating over it, so it feels like a hidden demo. Top video: a creator panics after autocorrect sent 'I love you' to her crush, and Bro suggests 'Autocorrect hates me, lol'. Simple 'fix my mistake' stories beat more complex ones (e.g. how to rizz a guy). They mass-produced videos because they could not predict winners.

After the mock-up video validated the format, they searched TikTok hashtags matching the scenario (e.g. relationship), looked for creators like those they found, and emailed them to start posting the proven format and variations. Included creators outside the US (e.g. Norway). Outreach was easy; getting the content to work took far longer and was a lot of management work, so Roger asked the host for help finding people to manage the campaigns.

'Group Picture' inside Lobby's video hangouts: anyone presses a button, everyone sees a 3-2-1 countdown and their phone vibrates, and the app captures a boomerang of the whole group. The output is watermarked and gets shared to Snapchat and Instagram Stories. The button lights up every 5-10 minutes to hint it's time for a picture, and leaving a chat prompts a goodbye picture. It must be a shared moment everyone knows is coming (posing, silly faces), not a secret screenshot. Two effects: more sharing out, and friends who weren't in the hangout feel left out and join next time.

Comparing Israel (viral, #1) with other markets showed one key difference: network density. They tracked the percentage of new users who got 5 friends on day 1 (~60% in Israel vs ~20% elsewhere), which went with D90 retention of 25-30% vs about half that. The conclusion was to raise density 'whatever it takes' through growth features where users invite friends by using the product, not content marketing (which brings random people without friends). They shipped the common invite and sharing mechanics plus unusual ones (Group Picture, one-minute roll call). Content still seeds the first users, but product-led invites scale it.

A game layered on the real-time product to fix 'temporal density' (friends must be online in the same second). At a random minute each night a bar pops up: 'talk to your friends for 1 minute, right now'. Friends join, catch up, and the call ends automatically after 1 minute. The 1-minute cap makes it low pressure (nobody feels held hostage; 'blame the game'). Inspired by BeReal's synchronized daily prompt.

Before engineering, build a tappable mock-up in ProtoPie of any flow (from the lock screen, the core talking flow, the community feed) and carry it around, using it daily to feel what works ('like a brick shaped like the next phone'). Each session produces insights such as 'this flow is too long'. The bar is that the team should feel good using it, not that it looks good on TikTok. Figma screens often look fine but fail once tappable. After the mock-up, build the app, test with family and friends, then start TikTok UGC.