Scaling Meta Ads from zero to $100K a month — Ethan Ethier, Built With Science

Launched the app in January 2025 to the YouTube audience but ran no Meta ads until January 2026. Spent that year acting on launch feedback and improving trial-to-paid, retention and churn, reasoning that paid traffic converts much lower than warm organic traffic. Only once those numbers looked right did they add acquisition channels (Meta, SEO/AEO).

Converted a web course (spreadsheets for workouts and nutrition, ~100K members over the years) into an app after members kept requesting it. Hyped the launch to members, email lists and publicly on the ~3M-subscriber channel.

Content-first: most long-form YouTube videos include a smooth, soft transition to the app in the middle or near the end rather than a hard pitch (hard pitches only on Instagram shorts). Every click from the video and description link is tracked per video, which showed that stories about a user who transformed their body on the app drive more trial starts than broad educational videos (a diet or a workout program) that reach more people. New videos are now planned around stories that tie into the product pitch.

Recently onboarded a few clippers who cut existing long-form YouTube videos and post them on Instagram Reels and TikTok, run independently of Jeremy. Hooks from clips that go viral feed back into scripting future long-form videos.

Deliberately avoided relying on Jeremy (already stretched) to remove the single point of failure of the YouTube channel. Creatives are B-roll 'sell' videos, testimonial content and many AI-made static ads. Angles are built around specific avatars and their pain points, sourced from surveys and past warm-traffic customers; winners are iterated while new avatars are slowly added. Run in-house rather than via an agency. After the Andromeda update, the focus is on better creatives rather than campaign/bid/targeting tinkering in Ads Manager.

Expecting cold Meta traffic to convert trial-to-paid lower than warm organic traffic, they first ran ads to gather data on the paid trial-to-paid rate, then used that baseline to calculate how much they could spend to acquire a customer before scaling and testing angles.

Kept Meta in-house so someone passionate about the product writes ads that don't over-promise. Tested click-baity ads that cheaply drive clicks and trial starts, but judged them on trial-to-paid; those showed a big drop. Ads built around a specific person's pain points and how the app uniquely solves them brought higher-quality trials.

An engineer built a custom dashboard that validates each new trial as coming from Meta, because heavy organic reach makes Meta blend sources. Trial-to-paid is tracked per Meta trial and the highest-intent users are fed back to Meta as conversion signals.

About 90% of traffic goes to a long personalized quiz on the website. The quiz educates users on the product, overcomes objections, raises perceived value and shows how personalized the plan will be; answers are used to build the user's plan in the app. Tested traffic to the App Store vs the quiz: the quiz converted much higher. An almost identical quiz inside the iOS app converted worse than web, while the same quiz on Android converted about as well as web. They may retest App Store traffic later.

Priced at about $189/year or $30/month since launch and never changed. Positioned against the cost of a personal trainer ('a personal trainer in your pocket') rather than as a workout/nutrition tracker; Meta ads never call it a tracker but a coach guiding you. Trust from the YouTube channel supports the premium. Price testing deliberately postponed because price changes distort product metrics while they work on retention and trial-to-paid.

Prompted by members asking for a family plan. The web quiz asks 'Do you have a gym buddy? Would you like to add them onto your plan?' Both users get 15% off. Those who decline see only a small gym buddy toggle on the pricing page, keeping the paywall simple. Buddies can see each other's workouts and meals and send nudges in the app.

SEO from old blog posts once drove 30-40% of traffic but declined with AI search. In 2026 they hired an external AI-search consultant, optimized existing content, and built free web tools (calorie calculator, macro calculator, etc.) that give real value and transition into the web quiz. Tools tend to rank well in AI search.

1) Use Amplitude to find the biggest drop-offs with the most room to improve across web and app onboarding. 2) For the chosen page (e.g. pricing), list as many problems as possible (lack of trust, unclear value proposition) and agree as a team on the biggest one. 3) Only then list solutions (e.g. testimonials for trust) and run the test. Document everything instead of Slacking an engineer, so failed tests can be diagnosed and iterated. Test bigger swings batching several hypotheses, since traffic is too low for many small tests; strip down if a batch fails. Apply winning messaging from one part of the funnel (quiz) to others (app).

Previously announced every feature by email to all members, which inflated the engagement metrics they wanted to measure. Now: tell a beta pool (~100 long-time former course members plus ~1,000 beta testers) about the feature, collect feedback and iterate, then release it publicly with no announcement and watch the metrics. Only once there is signal it adds value do they announce it.

Jeremy AI: a chat trained on all of Jeremy's YouTube videos and blog history that answers fitness and nutrition questions as Jeremy would, and also reads the user's workout, nutrition and progress data to personalize answers. Built in-house by one engineer over a weekend.

Jeremy collaborates with lesser-known researchers, coaches and dieticians on videos about their findings, partners with UBC on studies, and runs studies inside the app with customers (e.g. a 3 vs 6 training days per week study, using in-app data). A team of dieticians, coaches and researchers watches for new research to inform content.

Before: monthly and annual shown side by side. Variant: only the annual plan shown up front, with a Blinkist-style timeline above it: Day 1 you get your personalized plan; Day 7 your plan adapts to you; Day 12, before the 14-day free trial ends, 'we'll send you a reminder' and you can cancel anytime. Monthly is reachable via a 'view all plans' link. Price and positioning unchanged; UI-only change.

To make the value proposition clearer, added more context/copy to one of the pricing pages as an A/B test.

Paywall toggle: pay upfront with no trial and get 20% off (with a 30-day money-back guarantee), or toggle off to take the 14-day free trial. The copy framed no-trial as much better and the trial as losing out. Expected a win because much traffic is warm and trusts the brand. Inspired by in-app paywalls; not seen on web.

After the no-trial toggle lost, the documented hypothesis was that the copy framed no-trial as far better and the trial as losing out. Added a new test arm where both options are framed as good choices, neither as bad.

Rented a full studio and brought in six or seven participants to measure body fat (DEXA), total energy expenditure and actual intake, over months of work, as a big bet on something new.