How Cal AI turned creators into a growth machine — Jake Castillo, Cal AI

Thesis: people who track calories most are in health and fitness, which has millions of creators but few brands to work with them (protein powder, athleisure), so many had never had a brand deal. These creators already posted 'what I eat in a day' / 'to look like this you have to eat like this' videos about once a week, so the ask was simply to scan their food with Cal AI on camera. Jake's test for whether this transfers: huge creator supply, few competing brands, and content that naturally fits the app (an astrology app fails all three, and its few creators may be locked into exclusivity deals).

Dawson Gibbs, one of the first creators contacted, said yes and was willing to post more, iterate and try different angles and formats without charging per extra request. Cal AI reciprocated with unprompted bonuses and a real friendship. Once a format worked with him, it was copied and pasted to other fitness creators.

Cal AI made a build of the app specifically for creators so that while they scanned food on camera, both the Cal AI name and the logo were visible.

Cal AI's demo was a three-beat 'magic moment': scan, analyze, result. The streak feature popped up right after the first scan of each day and added a fourth beat, so creators were told not to film their first scan of the day.

For creators whose integrations were subtle, videos carried an on-screen caption 'tracked with Cal AI' to drive the brand home. For creators whose integrations were already obvious, the caption was left off because it would hurt views. Part of constant ongoing testing of small details.

Thesis: UFC fighters track food meticulously when cutting weight, so a photo calorie tracker is useful and interesting to them and could open a new niche of 1,000 more creators. Cal AI signed a couple of big UFC creators.

Mukbang creators eat 10,000+ calories in one sitting; the idea was a 'wow factor' of showing how easily the app tracks all that food before they eat it.

Views and followers can lie; comments can't. Niche level: read the comments on the big creators in a space and check whether people are asking for a solution your app provides. Creator level: compare comment sections, e.g. a leg-day video whose comments are all men posting fire/sweating emojis versus one where viewers write 'thank you, I wasn't sure how to work my hamstrings'. The second kind converts. Look for a cult-like following, not just a following.

Judge content quality, not current views: a cult-like following is a lagging indicator. Creators without a track record can be signed on cheaper deals. Jake signed a funny, charismatic creator (roughly his 12th-15th signing) getting 10-12K views; kept a friendly relationship through two quiet months; in month three a video hit seven million views.

Set the rate so the creator's average views roughly break even or make a little profit (the most likely outcome). Real money comes from the occasional mega-viral video, and with a flat fee all of that upside is profit. Avoid per-view pricing where possible. Bonuses and scaled deals were also used on top.

Work out revenue per thousand views (RPM) within the niche. Early on, with no other marketing, measure it directly: sales on a day a creator posts vs zero sales on days nobody posts. Later, keep reading smaller spikes. Never pay a CPM above that RPM; then set each creator's rate from that profitable CPM times their expected views. Accept that many creators will say no.

Once 50-70 influencers posted per day and several went viral the same day, spikes could no longer be read one by one. Jake built a 'renewal score' combining many metrics (engagement such as comments/likes/saves alongside views), plus incrementality checks comparing the demographics of a creator's following with where the spike showed up.

Build personal relationships with creators, especially early when there are few. Pay discretionary bonuses ('you crushed it this month, here you go') even when the deal had no bonus. Handle creators you can't afford with kindness. Good treatment leads creators to refer other creators.

When a creator demands a number (e.g. $1,000 a month), they care less about deliverables once the cash is met: agree to the fee and raise the number of posts so the effective CPM stays profitable. Get on calls, be a regular person rather than a 'sleazy agent', and meet in the middle.

Using management-consulting habits, Jake built systems and automations in Google Sheets (his 'Magnum office') to track deals, posts and payments for hundreds of creators, since existing tools disappointed him. He later turned it into his own software.

Thousands of messages to creators, kept up every day (on the order of 100 a day) for 18 months. Most didn't reply; many who did asked for rates too high to afford, and every negotiation was hard. Jake's view: being willing to do this grind is the edge, and it is harder now because creator rates have risen with more brands competing.

A couple of months in, Cal AI started paid ads (Meta) to diversify risk and because paid takes less work per dollar than influencers. The first creative was recycled influencer content.

Organic videos swing for a viral hit with a subtle integration; paid pays for every view, so every view should make clear what the product is. Cal AI invested in dedicated paid creative with a hook related to the product (not a random scroll-stopper) and a quick, interesting value prop. Jake learned the craft from friends at Speechify.

Used Tribe, where creators make ad creative for free and get paid a percentage of the ad spend that goes behind their videos, aligning their pay with what works.

Keep messaging congruent from ad to app. In onboarding, reiterate the pain points the marketing hit ('you're here because this resonated'), drive home what that pain has caused, offer the easier way and what it helps achieve, then ask what solving it is worth, and present a modest price ($30 a year) after several minutes of onboarding.

The Cal AI team shared successes and failures on Twitter while building. This invited copycats, but Jake believed clones lacked the obsession behind many small compounding improvements. Viral threads about Cal AI clones' revenue kept pointing people to Cal AI.

Despite running many price experiments, co-founder Blake Anderson insisted on optimizing for mind share, not profit: the most popular app is the easiest to buy, which means the lowest price. About $30 a year vs MyFitnessPal's ~$70. Fit the audience: young users with little disposable income who haven't felt the pain long enough to pay much to fix it.

Public in-app affiliate offer: anyone (including creators who wanted a promo code) can sign up and earns $10 per paid signup, not per download. One-time payment, about a third of a ~$30 annual sale, so no lifetime renewal liabilities.