When Experts Speak:Sequential LLM-Bayesian Learning for Startup Success Prediction

2025-12-23

AI Summary

To address the challenge of startup evaluation in venture capital—exacerbated by information asymmetry and scarcity of quantitative data—this paper proposes a sequential LLM-Bayesian model grounded in expert phone interviews. The model dynamically disentangles semantic content and evaluative signals at the granularity of question-answer turns, formalizing expert dialogues as a temporal Bayesian updating process with automatic attenuation of contradictory judgments. It is the first to empirically demonstrate that conversational cues strongly predict success for technically complex, early-stage, diverse-team, and low-visibility startups. The architecture integrates LLM-based semantic parsing, sequential Bayesian inference, and attention mechanisms, with ablation studies confirming the efficacy of each component. Experiments show a 6.691% improvement in F1-score and a 15.255% increase in portfolio return on investment, significantly outperforming existing text-based screening methods.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Evaluating startups is inherently challenging in entrepreneurial finance, where investors confront severe information asymmetry and limited quantitative data. Leveraging a novel expert network call data, we develop an LLM-Bayesian model that analyzes these conversations at the question-answer turn level, extracting semantic and evaluative signals via large language models (LLMs) and aggregating them in a sequential Bayesian architecture. The model dynamically updates beliefs as additional expert calls occur and attenuates contradictory assessments, which are absent from existing text-based screening tools. Empirically, our model outperforms state-of-the-art benchmarks by 6.691% in F1-score and increases portfolio-level Return on Investment by 15.255%. Attention and ablation analyses reveal that conversational cues are particularly informative for technologically complex startups, young firms, diverse founding teams, and firms with low public visibility. By converting expert dialogue into continually updated probabilities, our model advances research in entrepreneurial finance and information systems and offers policy implications for improving funding outcomes for informationally disadvantaged startups.

Problem

• Predicting startup success using expert conversations
• Overcoming information asymmetry in entrepreneurial finance
• Updating investment beliefs with sequential Bayesian analysis

Innovation

• Sequential Bayesian architecture for dynamic belief updating
• LLM analysis of expert conversations at turn level
• Model attenuates contradictory assessments in startup evaluation