Flamme · B2C · couples app
Relaunch the date planner as an AI relationship coach right after ChatGPT came out
Day-30 retention rose from 0.8% to 4.2%, still too low; press but no traction
Mixedfeature-developmentretention
What they did
After a 0.8% D30 date-planning MVP (a once-in-a-while tool, like booking an Airbnb), relaunched in December 2022 as Flamme AI, an AI relationship coach (ChatGPT wrapper) website, riding ChatGPT's release; it was pitched as the first AI for relationships and got TechCrunch and LA Times coverage.
What happened
D30 retention 0.8% -> 4.2%; strong media, but still dismal retention, so they pivoted again.
This shipped together with other changes, so the result can't be credited to this alone.
“This was also coupled with the times, ChatGPT came out. So it was December 2022 when it was released, and we decided to leverage this opportunity to build an AI-first product, which was Flamme AI.”
“Like, even though we could get media presence, even though the retention went up to from 0.8 to 4.2, it was still dismal, and we realized we had to do something else.”
Related topics
Get tactics like this every Monday
New tactics from the week's founder interviews, each linked to where it was said.
More from this episode
- Flamme: Make the daily question for couples the main thing on the home screenDay-30 retention went from 4.2% to ~30%, then 50%, with streak reminders
- Flamme: Send a friendly morning reminder, then an evening 'your partner will be sad' nudgeDay-30 retention went from 4.2% to ~50% alongside the daily-question redesign
- Flamme: Hide your partner's answer until you answer the question yourselfCreated a back-and-forth loop credited with lifting retention; no separate number
- Flamme: Add a home-screen widget showing distance to your partner and days togetherWorked well, per the founder; copied from a competitor; no numbers
- Flamme: Launch alongside a TechCrunch Startup Battlefield feature800% organic growth in one month; stopped by a trademark letter
- Flamme: Start charging in a free app when the money runs outBarely survived; the early '15% conversion' counted renewals, so it was inflated
- Flamme: Pitch big investors on great retention before you have a way to get usersFailed; top firms asked about distribution and no term sheets came
- Flamme: Scroll TikTok 4 hours a day for 2 months to learn what worksGot the first account getting views, then copied the approach across accounts
- Flamme: Run 20-30 TikTok accounts on numbered iPhones, 3-4 accounts per phone~50M organic views; host cites ~150K users
- Flamme: Use one Apple ID on every farm phone to skip repaying for the proxy appFailed: TikTok linked the accounts, one flag hit all; lost 120 work hours
- Flamme: Under 300 views means a broken account; 300-1,000 means fix the contentTheir working rule for which accounts and formats to fix; no numbers on impact
- Flamme: Judge app videos by likes and question slideshows by shares and savesLow-like videos rarely drove downloads, even at 1M+ views
- Flamme: Post slideshows of '5 uncomfortable but healthy questions to ask your partner'Shared a lot, but only 0.05-0.2% of viewers downloaded
- Flamme: Post videos that show the app in use, not memesApp demos convert 0.5-1% of views vs 0.05-0.2% for memes and slideshows
- Flamme: Give AI a 10-page brief with product details, winning scripts and a master promptTeam posts ~200 pieces a day from it; winners are fed back in
- Flamme: Log every post by hand in a spreadsheet and review it every morningForces the team to study why posts fail; no numbers given
- Flamme: Copy only formats that went viral on small accounts and get 'what app?' commentsAdopted after 50M views brought few conversions; C part learned late
- Run real 'what do I text back?' Reddit screenshots through the app on videoJust started; guest expects 0.5-1% of viewers to download
- Design the app's screens and loading animation to look good on TikTok