The Reliability Inversion
When synthetic consumer panels built with off-the-shelf LLMs outperform human ones, why the math has flipped, and where the pattern breaks.

I launched my first project when I was eleven. An online community about natural sciences that grew to 15,000+ active users within a year, cool for just a kiddo. I did it because I was a lonely kid looking for friends.
I'm Italian. Born in a small town near Venice, currently splitting time between Italy and Barcelona. I'm based at Norrsken, so if you're there too, say hi ๐.
In the last 10 years I've worked as a growth advisor, launching dozens of direct-to-consumer brands across Europe and the US. I designed the launch strategy of a brand under the Moncler group, and contributed to building from scratch one of the most successful DTC brands in France in recent years.
I'm a nerd by default. I get obsessed with things that have nothing to do with my day job, go deep, and try to understand how they actually work. This past year I got into chess โ I'm genuinely terrible at it, but I love it anyway. Picked tennis back up and I'm playing pretty well again. I also gave myself a challenge: get back to playing piano. Still fighting with scales. As a side project, just for fun, I'm building a personal AI agent that tracks my blood tests, diet, and training, and puts together a weekly plan based on the data. A tiny AI doctor that takes care of my health. It started as something for me, but I hope it can grow into something useful for other people too (thinking to opensource it).
Growth Marketing Studio. We grow consumer brands across Europe and the US. Tens of millions in ad spend managed, clients like JNPR Spirits and the new venture from the founder of Palm Angels. Every week we launch and test hundreds of new ad creatives to cut acquisition costs and maximize return on ad spend. AI handles the repetitive work, the team focuses on strategy.
Visit websiteAI Creative Studio. We produce AI-generated video assets that no traditional production could create. Short storytelling pills that tell brand stories in ways that feel more engaging, more surprising, and impossible to ignore. The output is AI-native, not AI-assisted.
Visit websiteConsumer brands burn months and six figures on qualitative research before every big call. We're replacing that loop. We build synthetic consumer panels: LLM agents demographically conditioned on a brand's real audience segments, each calibrated with its own background, preferences, biases, and purchase behavior. Then we run the research. Concept tests, purchase intent surveys, positioning sprints. The agents respond with natural language, not canned ratings. We map their open-ended answers to validated scales using semantic similarity scoring, so the output is statistically comparable to real panel data. Early benchmarks show 85%+ distributional similarity to human responses. Same depth, fraction of the time and cost.
When synthetic consumer panels built with off-the-shelf LLMs outperform human ones, why the math has flipped, and where the pattern breaks.
Everything around us was built by people like us. Some feel the pull to keep making things. Most never do. That's the question I keep coming back to.