An AI brainstorm for 110,000 students

How did generative AI take Red Bull Basement from 4,500 to 110,000+ entries?

An AI brainstorming tool made the entry easier to finish, and Red Bull Basement went from 4,500 entries to more than 110,000, from 2,000+ universities. Basement asks university students around the world for startup ideas. Plenty of students had one, but far fewer got it written into an entry, so we built a tool, a language model with image generation, that helps a student put an idea into words in more than 10 languages. I wrote its product specs, made completion the measure, held the cost to $0.07 per student and negotiated Microsoft’s co-funding of the servers.

The slow step was the form

Students had ideas, but the form was long, writing a pitch is hard, and few of those who started an entry sent it. Industrial engineering taught me to look for the slowest step in a system and fix that first. Here the slowest step was a student staring at a blank page, so that’s where the tool went.

What I did

We built it in three months. It was Red Bull’s first large generative AI product, and my part was:

1. Writing the product specs for a brainstorming tool that pairs a language model with image generation.
2. Making completion the measure: how many students who start an entry finish it.
3. Costing each student’s whole path through the tool, and working with the team on the architecture that held it to $0.07.
4. Negotiating $70,000 of Azure co-funding from Microsoft for the servers.
5. Owning the roadmap, the API architecture and the scaling, and solving the server problems as entries grew.

Measured by who finishes

We judged the tool by finished entries. Of the students who started one, 32% finished it, and entries passed 110,000 without a bigger team behind the program.

Seven cents a student

A free tool for students at 2,000+ universities only works if each student is cheap to serve. So I costed the tool per student, from a first idea to a finished entry, and we built to that number: $0.07. Microsoft’s $70,000 of co-funding covered servers on Azure as the entries came in.

Mentoring the teams

I mentored student founders, made mentorship materials and a dashboard for the colleges taking part, and helped choose the teams for the global final in Tokyo. I also worked with Basement’s global lead on the program’s brand and strategy, so it spoke to student founders and still sounded like Red Bull.

What it taught me

AI doesn’t create growth. Reducing friction does. The model mattered less than where we put it: on the one step that stopped students from entering. It’s the same habit behind the video tagging pilot I led: find the slowest step first.