Cal AI’s UFC creator bet: no download spike

The creator fit. Did the viewer?

Cal AI paid UFC creators to promote its calorie tracker and saw no download spike, according to co-founder Jake Castillo in a RevenueCat interview article published on 14 October 2026. His explanation: the athletes’ interest in tracking food did not necessarily match why their audiences watched.

This is a founder’s retrospective, not an October campaign result. Exact campaign dates, country and platform splits, measurement windows and incremental installs are not supplied. No visible spike does not mean zero installs, and the account cannot establish that audience mismatch alone caused the outcome.

Our reading: a creator can plausibly use your app while their followers have little reason to install it. “This person belongs to our category” is a weaker acquisition hypothesis than “these viewers are trying to solve this problem right now.”

Listen for a job the viewer wants done.

Castillo describes checking comments for requests that the product could answer, and demonstrating Cal AI’s scan-to-result sequence inside a creator’s routine. Read the founder account. The official iPhone listing describes meal photos producing nutritional breakdowns; scanning analysis results require a subscription.

Our interpretation: comments can help write a testable promise. They are not a demand forecast. Comments may be selected, unrepresentative or inauthentic, and even a sincere request does not prove willingness to pay.

Hypothetical example: an iPhone guitar-practice app considers two creators. One posts spectacular solos; another helps beginners change chords without stopping. Both attract guitar fans. The second gives the team a more specific hypothesis: viewers who ask how to practice that transition might want a short guided exercise. The first could still work; the category alone cannot settle it.

Before negotiating a larger package, write the viewer’s unfinished task in one sentence. Then ask whether the app can actually help with it during the first session. If the answer needs a completely different product, the partnership is premature.

Buy one useful demonstration first.

For that hypothetical guitar app, propose one clearly disclosed paid video with a capped fee and production budget. Let the creator show a real exercise in the shipping iPhone app: choose the difficult transition, practice it, and see the feedback. Avoid promising an instant skill improvement that the product cannot deliver.

Keep the invitation consistent through the App Store page, onboarding and any subscription offer. Someone responding to a beginner exercise should be able to find that exercise, understand what is free and see what payment unlocks.

• Before posting: record the planned publication window, country, acquisition route, total cost and other campaigns running at the same time.
• After arrival: follow first-time iOS users who can be associated with the campaign. Count exercise completion within 24 hours and another completed exercise on a later day within seven days, using all those entrants as the denominator.
• Before renewing: wait for the same observation window to mature, review proceeds and costs, and record how many installs could not be attributed.

A creator link and a quieter campaign calendar help interpretation; neither proves incremental lift. Treat an uncontrolled pilot as exploratory. If the signal merits a larger budget, design a holdout or randomized invitation where feasible rather than declaring every attributed install an extra install.

Give the next deal a spending limit.

Hypothetical economics, not Cal AI results: 100 first-time US iPhone installs associated with one creator produce $300 in net purchase proceeds over days 0–30 after each install, inclusive. That figure already deducts store fees, applicable transaction taxes and refunds. Variable service costs are $50. The creator fee and production together cost $350.

$300 − $50 − $350 = −$100 through day 30, before other overhead. The cohort has not recovered acquisition costs in that window. Later renewals could change the result, but assumed lifetime value should not be booked as money already earned. Uncertain attribution also makes this a campaign accounting view, not an incremental-profit estimate.

The next decision can be small: revise the demonstrated task, try a better-matched audience, negotiate a lower total cost, or stop. A disappointing pilot is useful when it tells you which assumption to test next.

Use a few well-defined events to follow the promised action and a consistent value-versus-cost basis to price the next experiment.