An AI-Driven Predictive Framework Utilizing Alternative Textual Data, Founder Networks, and Macroeconomic Indicators for U.S. High-Growth Startup Identification

• Adaan Ahsun

  Covenant University

  Author

Early-stage venture capital allocation remains characterized by significant inefficiencies, with over 90% of startups failing within five years and investment decisions heavily influenced by network-based biases that reinforce geographic and demographic concentration. Traditional predictive models relying exclusively on structured financial indicators have demonstrated moderate accuracy (approximately 70–75%) while failing to capture the qualitative dimensions and dynamic network effects that drive entrepreneurial success. This research addresses these limitations by developing and validating a hybrid AI-driven predictive framework that integrates alternative textual data (company descriptions, founder narratives), founder network characteristics (investor connections, co-portfolio relationships), and macroeconomic indicators to identify high-growth U.S. startups. Drawing on a retrospective analysis of 47,000+ U.S. companies from Crunchbase data (2015–2025), the proposed framework employs a stacking ensemble architecture combining BERT-based language models for textual analysis, Graph Neural Networks for network representation learning, and gradient-boosted models for structured features. The framework achieves 89.4% accuracy in predicting 5-year success outcomes (exit via acquisition or IPO), significantly outperforming traditional structured-only models (76.2% accuracy). Feature importance analysis reveals that founder network centrality (β=0.37), textual sentiment coherence (β=0.28), and sector-specific funding momentum (β=0.22) constitute the most influential predictors. Simulation experiments demonstrate that framework-guided capital allocation strategies achieve up to 47% higher investment success rates than historical real-world investment decisions. This research contributes a replicable, open-source methodology for de-biasing early-stage venture capital allocation while providing actionable decision-support tools for investors, policymakers, and ecosystem stakeholders.

• PDF

Copyright (c) 2026 Adaan Ahsun (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.