<p>High-Growth Institutional Entrepreneurship in Africa: What AI Does and Does Not Tell You</p>

High-growth ventures remain understudied in Africa due to persistent data scarcity. This paper analyses the co-evolution of high-growth ventures and entrepreneurial ecosystems across the African continent. We introduce a scalable methodology that leverages retrieval-augmented generation (RAG) and large language models (LLMs) to extract founder-level insights from unstructured online data, producing AI-generated founder narratives. Recent advancements in LLMs and RAG enable the systematic synthesis of qualitative data into structured, comparable narratives at a scale and speed unattainable by traditional qualitative methods. To address the limitations of AI, we supplement these narratives with primary founder interviews, identifying blind spots and defining the boundaries of effective human-AI collaboration. Our analysis reveals that AI-generated narratives, may contain factual errors, framing errors and overlook private considerations that founders do not disclose publicly. We argue that AI-assisted qualitative research in data-sparse contexts still requires primary triangulation. Furthermore, we formalize the distinction between source-faithfulness and ground truth-faithfulness (from the founder's perspective) as a critical evaluative framework for AI-assisted research. Our findings also reveal that African unicorn founders not only scale high-growth ventures but also actively construct institutions: "high-growth institutional entrepreneurship".