DealsFlow: An Evidence-Gated Agentic Harness for Venture Capital Deal Triage

📄 arXiv: 2610.32380v1 📥 PDF

作者: Ming Ze Tang, David Cutting, Boyd Carson, Vasiliki Carson

分类: cs.AI

发布日期: 2026-10-23

备注: 13 pages, 15 tables, 2 figures. Submitted to the AI4F 2026 workshop (LLMs and Generative AI for Finance) at ICAIF 2026


💡 一句话要点

提出DealsFlow以解决风险投资评分不准确问题

🎯 匹配领域: 支柱九:具身大模型 (Embodied Foundation Models)

关键词: 风险投资 大型语言模型 人机协作 证据引用 评分机制 投资决策 效率提升

📋 核心要点

  1. 现有方法在使用大型语言模型进行投资评分时,容易出现评分膨胀和难以审计的问题。
  2. DealsFlow通过人机协作的方式,结合LLM的能力和证据引用,确保评分的准确性和可审计性。
  3. 在243个投资案例中,DealsFlow的评分表现优于无约束LLM,且分析师的工作效率显著提高。

📝 摘要(中文)

大型语言模型(LLMs)能够阅读创业公司的投资计划书,但直接让其生成投资评分会导致评分膨胀、运行间差异和难以审计的理由。本文介绍了DealsFlow,这是一个在英国早期基金中部署的人机协作代理工具。该工具通过LLM读取计划书,引用证据并对每个类别进行评分;代码结合评分,设定上限并仅降低评分。通过三个机制实现这一点:对没有外部证据的事实类别设定阶段感知的证据上限(78、70或60分);怀疑者调用可能降低但绝不提高评分;以及在单次生成过程中预先获取和索引所有证据。我们在243个SEIS/EIS投资案例上进行了评估,结果显示,完整流程的评分与实际结果的Spearman相关系数为0.45,AUC为0.71,优于无约束LLM的0.38和0.68。

🔬 方法详解

问题定义:本文旨在解决使用大型语言模型进行风险投资评分时,评分膨胀、运行间差异和审计困难等问题。现有方法缺乏有效的证据支持,导致评分不可靠。

核心思路:DealsFlow通过引入人机协作机制,结合LLM的文本理解能力与证据引用,确保评分的准确性和可审计性。设计上,系统只允许降低评分,避免了评分的随意提高。

技术框架:DealsFlow的整体架构包括三个主要模块:1) LLM读取投资计划书并生成初步评分;2) 证据引用与评分结合,设定上限;3) 通过怀疑者调用机制进行评分的审查与调整。

关键创新:DealsFlow的创新在于引入了阶段感知的证据上限和怀疑者调用机制,这与传统的LLM评分方法有本质区别,后者往往缺乏有效的评分控制。

关键设计:系统设定了不同阶段的证据上限(78、70或60分),并在评分过程中实施怀疑者调用,确保评分仅能降低而不能提高。此外,系统在生成评分前预先获取和索引所有证据,以增强评分的可靠性。

🖼️ 关键图片

fig_0

📊 实验亮点

在243个SEIS/EIS投资案例中,DealsFlow的评分与实际结果的Spearman相关系数为0.45,AUC为0.71,显著优于无约束LLM的0.38和0.68。分析师在处理申请时的时间从44.7分钟减少到18.9分钟,且84%的引用声明得到了充分支持。

🎯 应用场景

DealsFlow的潜在应用领域包括风险投资、创业公司评估和金融决策支持。通过提高投资评分的准确性和可审计性,该系统能够帮助投资者做出更明智的决策,降低投资风险。未来,DealsFlow还可以扩展到其他领域,如信贷评估和市场分析,具有广泛的实际价值。

📄 摘要(原文)

Large language models (LLMs) can read startup pitch decks, but letting them emit investment scores directly invites score inflation, run-to-run variance and rationales that are hard to audit. We describe DealsFlow, a human-in-the-loop agentic harness deployed at a UK early-stage fund. The LLM reads the deck, cites evidence and rates each category; code combines the ratings, caps them and only ever lowers them. Three mechanisms implement this: a stage-aware corroboration ceiling (78, 70 or 60 out of 100) on factual categories without external evidence; a sceptic call that may lower but never raise a score; and a pre-fetch pattern that retrieves and indexes all evidence before a single generation pass. We evaluate on $N=243$ SEIS/EIS investments with recorded outcomes. Because recent investments have unresolved outcomes, we use the 142 that are at least three years old. There, the full pipeline's score has Spearman $ρ=0.45$ (95% CI 0.32 to 0.57) with the total-value multiple and AUC 0.71 for identifying write-offs, against $ρ=0.38$ and AUC 0.68 for an unconstrained LLM. The gain is small and comes mostly from the sceptic call; we did not test whether its lower-only rule matters. Investments that later exceeded $2\times$ had 64% of factual categories corroborated, against 38% for write-offs (omnibus Kruskal-Wallis test over five outcome groups, $p=0.031$), and our retrieval was not date-restricted, so these figures may overstate screening skill. In a study of 20 inbound applications, analyst time per application fell from 44.7 to 18.9 minutes including verification and correction, and 84% of 240 cited claims were fully supported by their sources, but 70% of reports needed at least one correction.