AI Outperforms VCs in Predicting Startup Success, Study Finds
Oxford and Vela researchers have developed VCBench, a groundbreaking benchmark demonstrating that large language models (LLMs) like GPT-4o and DeepSeek-V3 surpass top venture capitalists in identifying promising startups. The study analyzed 9,000 anonymized founder profiles, with AI models accurately predicting success—defined as exits or IPOs—where human investors often rely on biased heuristics.
The implications for early-stage investing are profound. If AI can systematically improve hit rates, it may democratize access to capital by prioritizing merit over network effects. This technological edge could soon become standard in deal-sourcing workflows across Silicon Valley and global venture ecosystems.