Internal First, Portfolio Second

先内后外：私募股权机构的 AI 战略顺序

The sequencing mistake most PE firms make with AI programs. And the 90-day internal win that changes the calculus. Blake Aber · Predicate Ventures · 2026 这是大多数私募股权（PE）机构在 AI 项目上犯的顺序错误，以及一个能在 90 天内改变局面的内部成功案例。Blake Aber · Predicate Ventures · 2026

The pattern I keep watching: a PE firm announces an AI initiative for its portfolio companies. The announcement is LP-facing. The initiative is visible. The firm looks like it’s ahead of the curve. Eighteen months later, the portfolio companies are still running pilots. The AI initiative is still being announced. The GP team has spent considerable time running workshops and writing frameworks. Portco results have been uneven. The problem isn’t that the firm was wrong about AI. It’s that the sequence was wrong. 我一直在观察这样一种模式：一家 PE 机构宣布为其被投企业（Portco）推出 AI 计划。这一公告是面向有限合伙人（LP）的，该计划清晰可见，让机构看起来走在时代前沿。然而 18 个月后，被投企业仍处于试点阶段，AI 计划依然停留在“宣布”层面。普通合伙人（GP）团队花费了大量时间举办研讨会和撰写框架，但被投企业的成果却参差不齐。问题不在于机构对 AI 的判断有误，而在于执行顺序错了。

Why firms start with portfolio AI 为什么机构倾向于从被投企业 AI 化入手

The LP optics pull strongly toward portco visibility. AI transformation at the portfolio level is a thing you can describe in a fund letter. It sounds like value creation. It’s what other GPs are talking about at conferences. Internal AI (AI that makes the GP team’s work better) is invisible from the outside. No LP ever asked a GP: “How much faster are you doing due diligence now that you’ve automated memo synthesis?” The incentive to publicize what’s visible outweighs the incentive to build what’s useful. So firms start with portfolio AI. The GP team hires a “Head of AI” or brings in consultants. The portcos get workshops. Some pilots run. Some don’t. LP 的视角强烈倾向于被投企业的可见性。在基金信中，被投层面的 AI 转型是可以描述的，听起来就像是价值创造，也是其他 GP 在会议上谈论的话题。而内部 AI（提升 GP 团队工作效率的 AI）从外部看是不可见的。从来没有 LP 会问 GP：“既然你已经实现了备忘录合成自动化，你的尽职调查速度快了多少？”宣传可见成果的动力超过了构建实用工具的动力。因此，机构从被投企业 AI 化开始，GP 团队聘请“AI 负责人”或引入顾问，为被投企业举办研讨会。有些试点成功了，有些则没有。

The problem with portfolio AI first 先做被投企业 AI 化的弊端

Here’s what the firms that sequence this way discover at around year two: they’re trying to provide AI support to portfolio companies without having built the AI muscle internally. When a portco CTO calls to ask for advice on eval harness design, the GP team doesn’t have a practiced answer. They haven’t built an eval harness themselves, so they can’t speak to why production AI is mostly harness and only a little model. When a portco initiative stalls at the organizational seam between product and engineering, the operating partner doesn’t know how to diagnose it. They haven’t worked that seam internally. The advice is conceptually correct but operationally thin. The portco can tell. The relationship suffers. 采取这种顺序的机构在两年左右会发现：他们在自身尚未建立 AI 能力的情况下，试图为被投企业提供 AI 支持。当被投企业的 CTO 打电话咨询评估框架（eval harness）设计建议时，GP 团队拿不出实操经验。因为他们自己没做过，所以无法解释为什么生产环境中的 AI 主要是框架而非模型。当被投企业的项目在产品与工程部门的衔接处停滞时，运营合伙人不知道如何诊断，因为他们内部从未处理过这种衔接问题。这些建议在概念上是正确的，但在操作上却很空洞。被投企业能看出来，双方关系因此受损。

Firms that skip internal AI first are usually still doing portfolio AI three years later with thin results. They haven’t compounded, because you can’t compound from a base you don’t actually have. 那些跳过“先内后外”步骤的机构，通常在三年后依然在做被投企业 AI 化，但成果寥寥。他们没有实现复利效应，因为你无法在没有实际基础的情况下实现增长。

The sequence that works 行之有效的顺序

Internal ops AI first (0-9 months): Start with the GP team’s own workflows. Three of them work in 90 days or less, without enterprise-scale infrastructure: 先做内部运营 AI 化（0-9 个月）：从 GP 团队自身的工作流开始。以下三项工作可在 90 天内完成，且无需企业级基础设施：

Due diligence memo synthesis. A named analyst spends 8-12 hours per deal synthesizing interview notes, financials, and reference calls into a memo. AI reduces this to a 90-minute review and editing job. Ships in 30 days. Immediate compounding.
尽职调查备忘录合成。分析师每处理一个项目需花费 8-12 小时来整理访谈记录、财务数据和参考电话，并撰写备忘录。AI 可将其缩减为 90 分钟的审阅和编辑工作。30 天内上线，即刻产生复利。
LP reporting automation. The quarterly LP letter has a predictable structure and a predictable set of inputs. AI drafts 80% of it. The senior team reviews and personalizes the 20% that matters. Ships in 45 days.
LP 报告自动化。季度 LP 信件具有可预测的结构和输入集。AI 可起草 80% 的内容，高级团队只需审阅并个性化处理那 20% 的关键部分。45 天内上线。
Deal-screening signal surfacing. The analyst team is reading a thousand news items and LinkedIn posts a week looking for signals that match the investment thesis. AI surfaces and scores the relevant ones. Human judgment remains the gate. Ships in 60 days.
项目筛选信号挖掘。分析师团队每周阅读上千条新闻和 LinkedIn 帖子以寻找符合投资逻辑的信号。AI 可自动筛选并为相关信息评分，人类判断依然是最终把关人。60 天内上线。

Internal investment AI (9-18 months): Build on the internal muscle. Now the GP team can talk about AI from experience, not from frameworks. 内部投资 AI 化（9-18 个月）：在内部能力基础上进一步构建。此时，GP 团队可以基于经验而非框架来谈论 AI。

Portfolio AI support (18+ months): Now you’re actually helpful to portcos. You’ve solved the problems they’re facing. You can speak to the organizational seam question, the eval harness question, and the build-versus-shepherd staffing question that sinks junior-IC-led programs from lived experience. 被投企业 AI 支持（18 个月以上）：此时你才能真正帮助到被投企业。你已经解决了他们面临的问题。你可以基于亲身经验，谈论组织衔接问题、评估框架问题，以及那些导致初级投资委员会主导项目失败的“自建还是外包”的人员配置问题。

The question worth asking 值得一问的问题

What is the internal AI your GP team runs on today? Not what you’ve recommended to portcos. Not what you’ve told LPs. What is the AI that your analysts and partners actually use, every week, to do their jobs faster and better? If the answer is vague, that’s where the AI program actually starts. 你的 GP 团队今天运行的内部 AI 是什么？不是你推荐给被投企业的，也不是你告诉 LP 的。而是你的分析师和合伙人每周都在实际使用、用以更高效工作的 AI 是什么？如果答案很模糊，那才是你的 AI 项目真正应该开始的地方。