We’re joined by Félix Boudreau, VP of Growth and Marketing at Pok Pok, the award-winning digital toy for kids. With years of experience scaling companies like Ipnos and WatchMojo, Félix is a proven growth leader who has driven major international expansion and industry partnerships.

---

About Felix: LinkedIn
About Pok Pok: Download the app

ABOUT ROCKETSHIP HQ: Website | LinkedIn | Newsletter | Youtube | Podcast Website

FULL TRANSCRIPT BELOW
Shamanth Rao:
I’m very excited to welcome Felix Boudreau to Intelligent Artifice. Felix, welcome to the show.

Félix Boudreau:
It’s great to be here, Shaman. I’m really excited to talk more about AI and apps.

Shamanth Rao:
I’m excited to have you, Felix, because I’ve followed your work for a very long time. I also admire you because you are one of the few people who actually walks the walk. There are a lot of gurus out in the world, and you have actually done the work, which I truly admire.

You’re not just a creator, but you’ve unlocked significant scale with creative events, and you’ll be kind enough to share that today, which I’m hoping you’ll continue to do in this interview. For all of those reasons, I’m excited to have you on the show.

Again, let’s start with the gurus. We, you and I, have all seen the posts that talk about end-to-end automation using AI: press a button and the outcome is a winner that generates millions of dollars. What’s your experience with these “push-a-button, get-a-miracle” solutions?

Félix Boudreau:
I have a lot of thoughts about that because I see those posts all the time on LinkedIn or X, from people claiming they create 500,000 videos every day or week. My question is always: how are they doing that? They often don’t include any performance data or screenshots of how it’s scaling campaigns.

I don’t disagree that you can create a lot with AI quickly, but in most cases, the real bottleneck is testing. Testing ads is very expensive. The problem isn’t creating 500 or 1,000 assets—it’s creating assets you truly believe will work and have strong faith in.

For us, we’ve experimented with creating end-to-end AI funnels with agents: doing research, turning that research into scripts, then into video. Most of the time, it’s very time-consuming and requires so many tweaks that the results are often underwhelming. In my experience, it hasn’t worked great.

The real impact of AI, especially in creative production, is empowering a team of individuals to produce more and add bandwidth. That can mean more efficient research by uploading thousands of reviews, distilling key insights, analyzing competition, or creating static assets quickly.

But it’s not a magical solution where you can go on vacation for two weeks while campaigns run automatically with awesome ads generated from scratch.

Shamanth Rao:
I completely agree. Anything we’ve tried still requires manual quality checks and intervention.

It reminds me of a quote I shared in my newsletter: there’s no greater waste of time than doing efficiently what shouldn’t be done at all. Every time I read these posts, I think about that.

If creative testing weren’t so expensive, the conversation might be different. Because it is expensive, the emphasis has to be on creating creative that is strong and connects. AI can help, but often human input is the more cost-effective solution.

Félix Boudreau:
Absolutely. Another underrated aspect is generating better ideas. There’s a lot of emphasis on “do more” but not enough on “do better,” which is what we’re aiming for.

You have to have intent in your creative process. The challenge isn’t producing 500 ads—it’s producing ads you understand and can double down on. Great hypotheses, ideation, iteration, and structured communication are key. Without it, AI can become “throw spaghetti at the wall,” which is hard to recover from.

Shamanth Rao:
Exactly. Having a human angle and clear intent is critical. I want to learn from your process, especially as creative output scales. One of our goals is to better synthesize learnings.

Talk us through your weekly creative cycle. What does your creative testing look like?

Félix Boudreau:
We have an unorthodox approach. From Thursday to Sunday, we test new creative. Each ad set has a hypothesis or concept, with 3–10 iterations. It’s structured: we know what we want to accomplish with each ad set.

We run it for four days. Poor performers may be turned off early to save money. Good performers might run longer and scale in budget. This is our testing unit over the weekend.

Monday is our creative meeting—the anchor of the week. Our creative UA manager analyzes all testing data, ranks ad sets by performance, and visualizes it in Figma. We dig into iterations: what worked? The hook, the image, the color?

Our team—UA managers, designers, myself, and sometimes guest contributors—brainstorms. We decide whether to push concepts further, iterate more, or kill them. If performance is terrible, we kill it.

Monday to Thursday is asset creation: designers prepare assets for the next week’s testing. That’s our weekly cycle.

Shamanth Rao:
That sounds comprehensive. You mentioned generating hypotheses. How do you do that at scale?

Félix Boudreau:
Currently, we run 7–10 ad sets, each with its own concept or hypothesis. Each ad set contains 3–10 creatives, mostly iterative. We aim to scale to 20 ad sets—150–200 ads.

The key is intent: when creating an ad set, you need a rationale. Is it based on user interviews, competitive research, or TikTok trends? Otherwise, discussing results becomes hard. Verbalizing why an ad should work helps the team align.

Shamanth Rao:
So articulating why you’re doing something is crucial. Seven to ten ad sets, each tied to a hypothesis, with variations. Do you get granular with the variations?

Félix Boudreau:
Yes, but there’s a balance. Conceptually, a meaningful difference defines a new concept; small changes, like background color, are iterations. For example, PPO is an educational app for kids 2–8, designed to keep kids calm.

We found success with ads targeting parents on planes. A new concept involved using the safety pamphlet aesthetic—completely different from previous in-plane UGC content. We iterate seven times on copy or visuals within that ad set. That’s one hypothesis: building on the parent-on-a-plane experience.

We structure experiments in two steps: first, copy variations with the same visual, then visual variations once we know which copy works.

Shamanth Rao:
So week one: test the pamphlet hypothesis with multiple copy versions. Week two: scale successful copy with new visuals.

Félix Boudreau:
Exactly. All tracked in Figma—a visual iteration network. Sometimes it takes four or five iterations to hit a winner. Structure helps dig into signals and metrics.

Shamanth Rao:
Right. You have infinite variations; intuition guides which to test.

Félix Boudreau:
Simplification is necessary. Testing minute differences too much wastes time and reduces impact. Go bigger and ensure you can interpret results.

Shamanth Rao:
And how do you manage multiple ad sets on limited budget?

Félix Boudreau:
We run 7–10 ad sets at a time. We aim for a 70/30 split: 70% iterating on winners, 30% new concepts. Iterating on winners is cheaper and often more effective.

We avoid bundling multiple hypotheses into a single ad set. Meta’s optimization would favor the most efficient concept, leaving others underfunded. Granularity provides insights into which concepts truly perform.

Shamanth Rao:
Makes sense. What was the inflection point to go from unstructured testing to this more granular system?

Félix Boudreau:
Two things: working with Jessica Bgo, a Berlin-based consultant, introduced structure; and attribution improvements from web-to-app allowed faster data, making smaller ad sets feasible.

Before, we relied on SKAN, which delayed feedback, making creative testing expensive and slow. Weekly cycles, visual boards, and structured processes improved velocity and knowledge management.

Shamanth Rao:
And the outcome inflection point?

Félix Boudreau:
After a few weeks of structured testing, we hit a big creative win. Iterating on it led to additional winners. Structured, intentional testing allows scaling winning creatives—using them across planes, cars, trains, etc.

Shamanth Rao:
That’s fascinating. Any examples of small changes with disproportionate impact?

Félix Boudreau:
Yes. A UGC video of a mom in a car initially performed lukewarm. By adding a fake tweet border and testing ten copy variations, one version—just a few extra words—blew up globally. Minute differences matter, hence multiple iterations.

Shamanth Rao:
How granular is your naming convention?

Félix Boudreau:
Very granular—a unique code for each iteration, consistent across platforms. This allows cross-platform analysis. We include iteration number, UGC name, ad length, etc., to parse data effectively.

Shamanth Rao:
And you track metrics in Figma or Google Sheets?

Félix Boudreau:
Metrics are mostly in Sheets. Figma links creative with key metrics visually. This hybrid setup balances speed and insight. Automation is challenging because processes evolve quickly, but Python has helped move things efficiently.

Shamanth Rao:
Fantastic. Felix, we’re close to time. For listeners, where can they find you?

Félix Boudreau:

Sure. I like to post a lot on LinkedIn lately, so you can find me there—Felix Boudreau. Otherwise, download Pok Pok. It’s a great app for kids, ages two to eight. You should check it out. Also, I’ll be at the Revenue Cat Up Conference in New York on October 14. So, if you’re around, come by. I think it’s going to be a really great event.

Shamanth Rao:

Wonderful. We’ll link to all of that in the show notes. And for now, this is a good place for us to wrap up. Thank you so much, Félix.

Félix Boudreau:

Thank you, Shamanth. That was a really interesting discussion.