起步顺序Order of work 先做机构记忆与检索,再做投后运营,最后监测 AI 敞口。Start with memory and search, then post-investment operations, then watch the AI exposure.
公开事实Public facts 已核对Checked
- US$50M2026 年 9 月投资法律 AI 公司 HarveyInvestment in Harvey, legal AI, Sept 2026属于一轮 6 亿美元融资Part of a US$600M round
- US$1.2B2026 年 3 月参与的 Wayve D 轮Wayve Series D round it joined, March 2026自动驾驶具身 AI 公司Self-driving embodied-AI company
- 346,000服务的在职与退休教师Working and retired teachers served其成员,投资成果支撑他们的养老金Its members, whose pensions this work supports
- Oct 2025首席投资官称正把 AI 融入分析CIO says AI is being built into analysis还包括市场扫描与内部沟通Also market scanning and internal communications
内部的三个价值池Three pools of value inside 判断Judgment
- 机构记忆Institutional memory(几十年的投资决策,散落在文件与人脑中。Decades of investment decisions, spread across papers and people.)→ 给一个资产类别的历史文件建检索助手。Index one asset class's past papers with a search assistant.
- 投后价值Post-investment value(Portfolio Solutions 团队专注私募资产的运营价值。Portfolio Solutions works on operating value in private assets.)→ 选一家组合公司的一个可复制流程,再复制。Pick one repeatable workflow at a portfolio company and copy it.
- AI 敞口风险AI exposure risk(大规模 AI 与风投持仓需要估值与情景监测。Large AI and venture holdings need valuation and scenario watching.)→ AI 持仓的情景视图,人工签字。A scenario view of AI holdings, with human sign-off.
- 沉淀知识Capture the knowledge → 复制有效做法Copy what works → 检查敞口Check the exposure
专家组裁决Expert panel 依据公开观点的推演Inference from public views
Foteini Agrafioti
每项 AI 使用挂到投资目标;一页纸写清研究助手、Portfolio Solutions 与数据的负责人。Tie each AI use to an investing goal; one page on who owns the research assistant, Portfolio Solutions and data.
Erik Brynjolfsson
知识工具的价值来得晚;定好分析师工时与决策质量基线,并为流程再造出预算。Value from knowledge tools arrives late; baseline analyst hours and decision quality, and fund process change.
Ziad Obermeyer
研究助手优化的是它能检索到的内容;先定义什么是好的投资备忘录。A research copilot optimises what it can retrieve; define what a good investment memo is first.
Ethan Mollick
分析师很可能已在实验;给安全空间,并测出助手在哪里编造事实。Analysts are likely experimenting already; give a safe space and test where the copilot invents facts.
Andrew Ng
从一个资产类别的备忘录开始;三到六个月出数字,再复制。Start with one asset class's memos; a number in three to six months, then copy.
Thomas Davenport
Portfolio Solutions 让它成为商业模式问题:AI 在哪里提升回报,而不只是降成本?Portfolio Solutions makes this a business-model question: where does AI move returns, not just cost?
Andrew Ng:从一个资产类别开始,用数字说话。Start with one asset class and show numbers.
Thomas Davenport:单点试点会掩盖运营模式的变化;要看全盘。A single pilot hides the operating change.
→ 一个资产类别试点,并从第一个月起绘制全盘负责人图。One asset-class pilot, plus a plan-wide owner map from month one.
六位专家是谁、怎样思考、有什么铁律 →Who the six experts are, how they think, their rules →
三个目标Three priorities 建议Recommendation
建立机构记忆Build the institutional memory
覆盖历史投资文件的检索助手,答案注明出处。A search assistant over past investment papers, cited to source.
把 AI 带进组合运营Take AI into portfolio operations
先做一家组合公司的一个可复制流程,再推广。One repeatable workflow at a portfolio company, then copy.
监测 AI 敞口Watch the AI exposure
AI 持仓的情景视图;估值判断仍由人负责。Scenario views of AI holdings; valuation stays with people.
场景组合Scenarios 建议Recommendation
| 场景Scenario | 做法Approach | 状态Status |
|---|---|---|
| 投研 CopilotInvestment research copilot 提效Efficiency体验Experience | 基于历史文件作答并注明出处,由分析师核对。Answer questions from past papers with citations; an analyst checks. | 建议Proposed |
| 投后运营手册Post-investment playbook 提效Efficiency增收Revenue | 把一个已验证流程复制到多家组合公司。Copy one proven workflow across portfolio companies. | 建议 · 基于 Portfolio SolutionsProposed · builds on Portfolio Solutions |
| AI 持仓情景监测AI-holdings scenario monitor 降风险Risk | 对风投与 AI 持仓做情景分析,人工签字。Scenario analysis of venture and AI holdings, signed off by people. | 建议Proposed |
| 市场扫描工作流Market scanning workflow 提效Efficiency | 把首席投资官所述的扫描与分析用法变成共享工作流。Turn the CIO's stated scanning and analysis use into a shared workflow. | 建议 · 基于公开表态Proposed · builds on a public statement |
12 个月路线12-month roadmap 建议Recommendation
奠基Foundation
- 基线:分析师工时Baselines: analyst hours
- 模型清单与分级Model inventory and tiers
- 全盘负责人图Owner map across the plan
试点Pilot
- 一个资产类别的检索Search over one asset class
- 对照组测量Control group measured
- 失败边界测试记录Failure-edge test log
上线Go live
- 该类别 Copilot 上线Copilot live for that class
- 分析师核对引用Analyst checks citations
- 上线即监控Monitoring from day one
扩展Extend
- 第二个资产类别Second asset class
- 投后运营手册Post-investment playbook
- 年度重排优先级Annual re-ranking
治理与合规Governance 建议Recommendation
- 模型清单;与法律顾问确认 OSFI E-23 适用性Model inventory; confirm OSFI E-23 applicability with counsel
- Copilot 每个答案注明出处,由分析师核对Every copilot answer cites its source; an analyst checks
- 估值判断由人负责Valuation judgments stay with people
三层指标Three-layer metrics 建议Recommendation
| 采用Adoption | 每周活跃分析师数Weekly active analysts · 各团队提问数Questions asked per team |
|---|---|
| 过程Process | 带有效出处的答案占比Answers with a valid citation · 分析师核对通过率Analyst check pass rate · 查找先例用时Time to find a precedent |
| 结果Outcome | 对照基线的分析师工时Analyst hours against baseline · 决策质量评审Decision-quality review |
目标值由企业自定。Targets are the company's to set.
风险Risks 判断Judgment
流畅但错误的答案Fluent but wrong answers
Copilot 可能从旧文件里编造事实;要求出处并核对。A copilot can invent facts from old papers; cite and check.
知识锁在人身上Knowledge locked in people
资深判断可能从未成文;需要有意识地沉淀。Senior judgment may never reach a document; capture it deliberately.
试点没有负责人One pilot, no owner
没有业务负责人,好试点会停在一个团队的边界。Without a business owner, a good pilot stalls at the edge of one team.
前 90 天First 90 days 建议Recommendation
| 第 1–2 周Weeks 1–2 | 点名试点的业务负责人Name a business sponsor for the pilot |
|---|---|
| 第 3–4 周Weeks 3–4 | 选定资产类别;定基线Choose the asset class; fix baselines |
| 第 5–10 周Weeks 5–10 | 给文件建索引;风险分级初稿Index the papers; risk tiers drafted |
| 第 13 周Week 13 | 首次复盘First review |
我们可能错在:We could be wrong if: 分析师更信几十年的个人判断而非 Copilot,或旧的非结构化文件检索效果太差。analysts trust decades of personal judgment over a copilot, or retrieval is too weak on older, unstructured papers.
出处Sources
本页是基于公开资料的分析,与所述企业没有合作关系,不代表其官方立场或内部计划;企业数字多来自其自述或媒体转述;「专家组」是依据六位专家公开观点的推演,不是他们的原话,也不代表他们或其雇主的立场。资料整理日期:2026-11-14。This page is an analysis based on public information. It has no affiliation with the company and does not represent its official position or internal plans. Company figures come mostly from its own statements or media reports. The expert panel is an inference from six experts' public views, not their words, and does not represent them or their employers. Compiled 2026-11-14.
想给你的公司也做一份?Want one for your company?
同样的方法:公开事实、专属判断、12 个月路线和前 90 天。公司大小都可以聊,中文沟通。Same method: public facts, a company-specific judgment, a 12-month roadmap and the first 90 days. Any size of business; we can talk in Chinese or English.