AI Beats Human Investors 4 Out of 5 Times Picking Startups

Artificial intelligence just proved it can spot winning startups better than trained human experts, and the implications could democratize investment knowledge for millions of entrepreneurs and small investors.

Researchers from the University of Michigan, New York University, and Indiana University ran a forecasting tournament with 30 real technology startups launching on Kickstarter. They asked both AI models and 346 experienced managers to predict which ventures would raise the most money, then waited to see what actually happened.

Google's Gemini 2.5 Pro correctly predicted the winner in nearly four out of five matchups. The best human forecasters got it right only three out of five times. Several other AI models also beat the human experts by significant margins.

The study's design was bulletproof. All predictions happened before any outcomes were known, and the startups launched after the AI's training data cutoff, meaning the models couldn't cheat by remembering results they'd already seen.

Professor Felipe Csaszar, who led the research, discovered something even more surprising. When they combined human judgment with AI predictions, accuracy actually dropped. The "human in the loop" approach that companies assume makes AI safer turned out to make forecasts worse for this specific task.

The AI's advantage comes from processing thousands of factors simultaneously without fatigue or inconsistency. Humans might evaluate the same startup differently depending on their mood or what they saw yesterday. The AI doesn't have those limitations.

The Ripple Effect

This breakthrough could level the playing field for entrepreneurs everywhere. Right now, getting funding often depends on knowing the right investors or living in Silicon Valley. If AI can identify promising ventures as reliably as top venture capital firms, that knowledge becomes available to anyone with internet access.

Small business owners in rural communities could get the same quality analysis as startups pitching to elite Sand Hill Road investors. First time entrepreneurs without industry connections could understand their real chances and get actionable feedback to improve their pitches.

The researchers were careful to note their study focused on one specific task: comparing early stage Kickstarter projects. Whether AI performs as well evaluating later stage investments or complex business models remains an open question requiring more research.

But the core finding stands. For the first time, machines can match and exceed human judgment on strategic decisions that experts believed required uniquely human intuition and experience.

That doesn't mean venture capitalists are obsolete tomorrow. The best investors provide mentorship, open doors, and help startups navigate challenges after writing checks. But their claim to superior pattern recognition in picking winners just faced its strongest empirical challenge yet.