Cal AI：用评论验证需求、RPM封顶定价与“魔法时刻”测试

The logic was sound. UFC fighters track their food meticulously when they cut weight, so a calorie tracker that works from a photo saves them weighing everything, and the fight audience is enormous. Cal AI signed a couple of UFC creators and waited. "The UFC creators we signed posted and crickets," says Jake Castillo, Cal AI's co-founder and COO. "No download spike. Nothing." Asked whether these were big creators, he answers with the bill: "thousands and thousands of dollars spent."

Watch on YouTube

The explanation came from the audience data. "Is the average person watching? UFC is doing so on their couch as they're drinking beer and eating pizza with their friends," Jake says. "The last thing they want to think about is tracking their calories, because that's just like going to give them bad news." Mukbang creators, who eat 10,000 calories in a sitting, were the same mistake from the other side: viewers watching someone else's cheat day are planning their own, not counting it.

What replaced the on-paper logic was a method anyone can run from their phone. "Views can lie. Followers can lie, but comments cannot," Jake says. For a niche, read the comment sections of its biggest creators and look for people asking for a solution your app solves. For an individual creator, the comments separate the one in ten who converts from the nine who don't. Two women post the same leg-day routine; under one it is fire emojis from men, under the other it is "thank you so much for posting this video. I was really unsure of how to work out my hamstrings or quads or glutes." Jake's line on which one converts: "I think it's pretty obvious which one will."

The magic moment test

Before any of that, Cal AI had to pass a test of its own. "What we called it was the magic moment," Jake says. "It was like a one, two, three punch. You scan, analyse results." When a fitness creator making a what-I-eat-in-a-day video scanned a plate on camera, viewers did a double take, and the double take is what sold the app. "If your app doesn't have a magic moment, it doesn't lend itself well to a seamless integration because it's not going to cause that double take." An app that needs a paragraph of explanation should run paid ads with a hard sell instead, because "if it feels like an ad, it's not going to go viral. If it's not going to go viral, it's not going to be profitable."

The test shaped details most people copying the playbook never saw. A streaks pop-up fired after the first scan of the day and broke the three-beat rhythm, so "we would make sure our influencers didn't show their first scan of the day." Creators with subtle integrations got an on-screen caption naming the app; creators whose integration was already obvious didn't, because the caption cost views. A dedicated creator build put the logo on every screen. "No marketing program should ever feel like a finished product," Jake says.

The one number that caps what you can pay

The deal math starts with revenue per thousand views, which Jake says is easy to find early on: with nothing else running, a creator's post shows up as a spike in a flat download chart. "Once you get that number, it's like you can't pay like a CPM that is higher than that RPM. Otherwise you are guaranteed to not be profitable." Creators will push back. "They may say no, like that's not I want to get paid more. But it's like you need to understand there is no other no other alternative."

From there the structure follows the shape of the returns. Break-even is set by average views; the profit is in the outlier. "Where you make your real money is when they have like one video go mega viral. And like all that is profit," Jake says. "So whenever possible, I would try to do flat rate deals because it allows you to capture the asymmetry that is algorithmic content." The creator in the byline is the case for it, and for the bets Jake says the UFC miss never stopped them taking: signed somewhere around the 12th or 15th deal on the strength of his content, doing "like 10,000, 12,000 views" for two months, then "month three, he had a video get 7 million views and then like for like from that point forward, it was like never under 500,000 views." Jake's framing: "A cult like following is a lagging indicator. If you're a great creator. The cult like following will come." Sign before it does, and the price is lower.

At scale the same discipline became a system. Cal AI had "50 to 70 influencers posting on a day," which makes attribution murky, so Jake built what he calls a renewal score that weighs the secondary signals before anyone gets re-signed. He ran it from spreadsheets for a long time, and is releasing the software he eventually built in its place at the end of October.

Jake Castillo on why the Cal AI clones were never going to make it

Cal AI posted its wins and losses on Twitter as they happened, and clones followed. Jake wasn't worried. "We knew that if people were building a Carly clone just because they heard about our success, they were never going to make it," he says, Carly being the transcript's rendering of Cal AI. "The reason why we were successful was like, we were obsessed." He was dreaming about influencer deals, and the obsession produced "a bunch of little, small, incremental improvements in those small incremental improvements compound over time to be an insurmountable mountain that competition can't climb." The Twitter threads about clone revenue sent traffic to Cal AI anyway.

He is equally direct about people copying him. "The easiest way for an app founder to blow a ton of money is to copy what I did or copy what you saw on a podcast because like, my app is not your app." The creator thesis worked because health and fitness has more creators than any other niche and almost no brands for them to work with, and those creators were already filming the food Cal AI scans. An astrology app, he points out, has a few hundred creators and the rival astrology apps have probably already signed them to exclusivity. The price is the other thing not to copy. Cal AI sat among the cheapest calorie trackers at roughly $30 a year because, in co-founder Blake Anderson's framing, "we're not optimising for profit. We're optimising for mindshare," and because a young audience that hasn't lived with the problem for long won't pay much to solve it. Different audience, different answer.

In the full episode, Jake also explains why Cal AI's viral creator posts underperformed as paid ads until the team made dedicated creative, how a five-star hotel lobby with a rickety elevator explains funnel congruence, and what he can and can't say about the MyFitnessPal acquisition.