Built With Science · B2C · science-based fitness coaching app
Build an AI coach trained on the creator's videos that reads each user's own data
One of customers' favorite features, founder says; no numbers
What they did
What happened
- Effort
- One engineer, one weekend
In their words
Scaling Meta Ads from zero to $100K a month — Ethan Ethier, Built With Science
David BarnardWell, and speaking of product launches, I know this year you launched something called Jeremy AI. What is that and how did it go?
Ethan EthierYeah, so Jeremy AI has actually been one of our customer's favorite features, essentially-
David BarnardWhich is surprising because a lot of people screw up their AI role.
Ethan EthierYeah. During the whole launch of the app and the growth of all our features, we've never really promoted or done much with AI besides some of the backend stuff. But Jeremy AI is essentially an AI chat that's trained off of all of Jeremy's past YouTube videos, his blog history, and it's a source for users to get any fitness nutrition-based questions answered, just like Jeremy would be talking to them on the phone.
Ethan EthierBut I think what we've done to really increase the value of this is not only is it trained off of Jeremy's research and content, but it's also able to read the user's workout, nutrition, and progress data so it could actually give really, really personalized answers based off where they're at.
David BarnardThat's really cool. If you don't mind digging a little bit into that, what are you using on the backend for that? Just one of the off-the-shelf AI provider, big models?
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
- Spend a year fixing trial-to-paid and retention before buying any adsFounder says it lifted the floor for every channel; paid traffic later converted at 25%
- Hype the app launch to your YouTube viewers, course members and email listA launch spike, then a steady trickle; not a growth engine on its own
- Mention the app softly mid-video, and tell user transformation storiesStory videos bring more trials than broad educational ones; old videos give steady monthly trials
- Hire clippers to post long-video clips on Reels and TikTok, then reuse winning hooksSome clips go viral; their hooks now shape new long videos
- Scale Meta ads without the famous founder: B-roll, testimonials and AI image adsScaled from zero to ~$100K/month (host figure); 25% trial-to-paid
- Spend the first month of ads just learning your paid trial-to-paid rateTook about a month to get a baseline to set the allowed cost per customer
- Judge ads by how many trials pay, not by how cheap the trials areClickbait ads brought cheap trials but a huge drop in trial-to-paid
- Build a dashboard that confirms which trials came from Meta, and send Meta your best users
- Send ad clicks to a long personal quiz on your website, not the App StoreQuiz beat direct-to-store; now gets ~90% of traffic; same quiz in the iOS app did worse
- Sell the app as a trainer in your pocket, not a tracker, at $189 a yearKept $189/year since launch; 35-40% organic and ~25% Meta trial-to-paid
- Offer 15% off each if users add a workout buddy to their plan15% of trial starts add a buddy; higher order value and better retention
- Build free calorie and macro calculators that rank in AI search and lead to the quizWorking well, founder says; tools convert slightly better than blog posts
- List every reason a page fails before brainstorming fixes, and write each test down~70% of tests fail first time; written notes turn failures into later winners
- Release new features quietly, and announce only once usage proves themFounder says it gives cleaner reads than email-hyped launches; no numbers
- Team up with researchers and run studies with your own app usersFounder says it adds credibility; no numbers
- Show only the yearly plan first, with a day-by-day trial timeline and reminder promiseAnnual picks rose from ~60% to 75-80%, with no price change
- Add more explanation to the pricing page to make the value clearerHurt conversion; extra text overwhelmed buyers who were ready to start
- Offer 20% off to pay now, with a toggle back to the 14-day free trialQuite negative, even with a 30-day money-back guarantee
- Make both the free trial and the pay-now discount sound goodVery strong lift after a few weeks, vs the version that shamed the trial
- Film a months-long science experiment video with real people in a rented studioOne of the year's top videos; decent but not major app conversion
Tags: ai-coach, personalization, creator-persona