| Large Language Models (LLMs) exhibit broad capabilities, yet their alignment with domain experts is limited. This is especially relevant for Venture Capital (VC), where misjudgments carry significant financial consequences. Investors selectively fund only about 1% of startups evaluated, making expertise-aligned decision systems critical. This paper examines research on LLMs in VC and related investment contexts, organized by investment stage and assessed against three dimensions grounded in VC practice: (1) capturing social signals, (2) reasoning under uncertainty, and (3) maintaining up-to-date market knowledge. We find that no approach comprehensively captures all three dimensions, with social signals receiving the least attention. Moreover, while most studies produce a final investment decision, only half use deliberative reasoning, and fewer support it with external knowledge retrieval. These gaps reveal directions for aligning LLM-based approaches with VC practice. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.