Sunflower · B2C · addiction recovery app
Let the in-app AI coach follow up on its own and reply in the community
4M+ messages handled in a year; users say it saved their lives
Workedfeature-developmentretention
What they did
Started as a ChatGPT wrapper, then prompt optimization, then an agent ('Sam') with app-wide context and memory: proactively follows up (e.g. on payday), replies to posts in the community, runs prompted sessions. Risky conversations escalate to a clinical team under a VP of Therapy; biometrics and data erasure protect users from abusive partners. Next: integration with a teletherapy clinic (relapse scores to therapists, homework follow-ups).
What happened
Over 4 million messages in the last year; many users credit it with staying sober.
In their words
The Superwall Podcast · Koby Conrad - I speedran an ai app from 0 to $1m/yr in 10 months“So, first it was just a ChatGPT wrapper, and then we did some like fancy prompt optimizations, and then eventually like we turned it into an AI agent.”
“so he can like follow up with you when it knows that it's payday, and he can he will show up in the social media and he will like respond to the posts”
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
- Sunflower: Hire the biggest creator in your niche as a salaried team member with equity50-100K link clicks a month from one creator; founder calls it the core of go-to-market
- Function Health: Get the best-known expert in your field to be the face of the companyGiven as an example of the approach working; no numbers
- Rupa Health: Ask conferences what they'd give you for double their top sponsor packageFounder says it beats small test buys: better treatment and brand effect; no numbers
- Sunflower: Launch a bare-bones first version with three features, then keep improving itUgly v1 grew into the current product; no early numbers given
- Sunflower: Start free like Duolingo, then add a monthly paywall once AI costs pile upFree model failed: investors wanted revenue and AI costs hit hundreds of thousands
- Sunflower: Add a yearly plan and pre-select it on the paywallAdding annual brought some annual revenue; defaulting to it brought 'a lot more'
- Sunflower: Default to paying now, with a toggle to switch on the free trialMade more money than the default-annual paywall; no numbers
- Sunflower: Lower your prices when your users have little moneyLower prices made more money; no numbers given
- Sunflower: Before the paywall, promise to remind people before their free trial ends+48% revenue, +70% annual revenue, +46% trial conversions, +30% trial starts
- Sunflower: Build everything around the one simple thing users do every dayTimer is the most important feature; AI chat alone had poor retention
- Sunflower: Pitch investors with revenue growth and a future vision, not user countsFirst user-focused deck failed; the revenue-and-future deck raised post-YC
- Sunflower: Put in your own money, then ask people you made rich to invest$100K of own money plus ~$350K from his network after years of no funding