Predicting Business Angel Early-Stage Decision Making Using AI

Abstract

External funding is crucial for early-stage ventures, yet business angel decision-making remains subjective and resource-intensive. The Critical Factor Assessment (CFA), a validated eight-factor venture evaluation framework, has demonstrated superior predictive accuracy over investors' own decisions. However, full evaluation requires multiple trained evaluators and several days per assessment, limiting adoption at scale. This study investigates whether AI can overcome these constraints. Multiple Large Language Models (LLMs) were prompted to assign CFA scores to 600 transcribed Shark Tank pitches with known deal outcomes. Machine learning classification models trained on the LLM-generated CFA scores achieved 85.0% accuracy in predicting deal/no-deal outcomes. The top-performing model (GPT-4.1-mini) exhibited very strong correlation with trained human evaluators (Spearman's ρ = 0.909, p < .001), substantially exceeding mean human–human agreement (ρ = 0.465). The integration of AI-based feature extraction with a validated decision-making framework yielded a scalable, reliable approach to early-stage venture evaluation.