How Flame built 50% day-30 retention and 150k users from TikTok

Executive overview

Most consumer apps die from poor retention or broken distribution — Flame solved both, in sequence. Starting at 0.8% day-30 retention, the team rebuilt around a single daily habit loop and reached 50% — higher than TikTok's own 40%. With no money left, the founder spent two months studying TikTok full-time, then built a systematic content machine across 20–30 accounts.

The core insight: retention without distribution is a dead end — master both, in that order, before raising.

From 0.8% to 50% day-30 retention

• Original app (date planning) had 0.8% day-30 retention — a tool used once, not a habit
• Pivot to Daily Question feature (couples answer a shared prompt together) anchored the habit
• Framework: every retention event needs a trigger, a carrot, and a stick
• Carrot: partner's happiness — answering the question made your partner happy, not just you
• Stick: streak notifications warned users their partner would be "sad" if they missed the day
• Missing-out mechanic: you couldn't see your partner's answer until you submitted your own
• Widgets on the home and lock screen kept the app visible between sessions — harder to ignore than push notifications
• Day-30 retention benchmarks for context: X ~30%, Instagram ~35%, TikTok ~40%, Flame ~50%

Why retention alone wasn't enough

• High retention failed to close investor meetings — tier-1 VCs (a16z, Sequoia, General Catalyst) asked about distribution and got no answer
• The lesson: solve distribution before pitching, not after
• Running out of money forced monetisation; early 15% "conversion" figure was inflated by renewals mixed with new users
• True distribution remained unsolved through most of 2024; the team nearly shut down

Learning TikTok from scratch

• Founder spent 4 hours/day for 2 months scrolling TikTok before posting anything
• Key diagnostic: under 300 views = account problem; 300–1,000 = content problem; 1,000–5,000 = iterate same format until it breaks 100k
• TikTok at scale is a volume game — one great video isn't enough; you need repeatable formats across many accounts
• Operated 20–30 accounts across ~10 numbered iPhones, each running 3–4 accounts
• Each phone required a full factory reset, unique Apple ID, and a separate proxy — sharing an Apple ID across devices fingerprints all linked accounts together

The VSC framework for content selection

• V — Virality: find content that already went viral on accounts with under 5,000 followers and over 100k views; confirms the format works independently of audience size
• S — Scalability: the format must be reproducible at volume without manual re-recording; AI tools (Sora, Flux/NanoBanana Pro, influencer.ai) enable this
• C — Convertibility: check comments — if people aren't asking "what is that app?" the content generates views, not downloads; memes and slideshows convert at 0.05–0.2%; reaction + app demo at 0.3–0.5%; straight product demo at 0.5–1%
• Convertibility was the last element discovered and the most important; the team spent months on viral formats that never drove installs

Content production system

• Playbooks: 10-page PDFs fed to AI to generate scripts and image prompts at scale
  • Section A: product and brand context, target audience, core features
  • Section B: winning script references — curated examples of formats that already worked
  • Section C: master prompt defining output structure, slide format, image generation prompts
• New winning content is added back to the playbook after each iteration — the system compounds
• Manual tracking in a shared spreadsheet (not a tool) forces daily review of every piece of content and every account; the discipline surfaces why things fail, not just what succeeded
• 200 pieces of content published per day across all accounts; even at 0.5% conversion and no virality, that's ~2,000 downloads/week

Building the product to perform on TikTok

• Every animation, loader, and UI flow was designed to be visually striking in a vertical video context
• Principle borrowed from apps like Locket: build the product so it spreads natively on social
• Eye-catching UI in a demo video removes the need for a hook — the product sells itself

Asking for help effectively

• Start with a small ask; build trust before requesting significant time
• Always give something back first — share a specific insight, a useful skill, anything concrete
• The founder's first contribution to the community was sharing Flame's retention framework; this opened doors to people who already knew TikTok distribution cold