Work

Signal Conversion Efficiency: Generative AI, Human Judgment, and Capital Formation in Token-Based Crowdfunding

Year: 2026
Type: preprint
Abstract: Information asymmetries between entrepreneurs and financiers shape the efficiency of entrepreneurial finance markets. Drawing on the distinction between information-endowment asymmetry (unequal access to signals) and information-utilization asymmetry (unequal processing of signals), we explore whether generative artificial intelligence (AI) improves ''signal conversion efficiency''(SCE), the effective utilization of the information endowment. We engineer an AI agent that uses information from blockchain-based crowdfunding campaigns to generate investment recommendations and compare them with recommendations by human experts and a human-built recommendation algorithm. AI advice outperforms other signals statistically and economically: Including it as a signal for capital formation raises R-squared by up to 35\%, and a one-standard-deviation increase in AI recommendation strength is associated with 48\% higher funding amounts. Explainable AI using Shapley values corroborates AI's predictive power. Consistent with our SCE conjecture, AI-human advice congruence is positively associated with capital formation, while the direction of divergence matters: campaigns rated more favorably by humans than by AI raise less funding than those for which AI is more optimistic. The AI agent's predictive performance increases across generations of large language models evaluated on the same class of historical disclosures. Our findings highlight information utilization as a central mechanism through which generative AI supports investment decisions in entrepreneurial finance. JEL Codes: C45, G24, G32, K22, L26
Language: English
Open Access status: green
OpenAlex ID: W7226008336
Cites: 64
Cited by:
Related to: 10