ai-readiness

Anthropic ｜ 金融與投資 ｜ 原文 ｜ Apache-2.0

掃描投資組合公司，找出槓桿最高的 AI 機會並排出優先順序

---
name: ai-readiness
description: 掃描投資組合，找出槓桿最高的 AI 機會，並排序 operating partner 的時間該投在哪裡。讀入多家投資組合公司的季度更新與財報，找出每家公司的快速見效項目（quick win），再彙整成一份排好序的行動清單。用在季度投資組合檢討、年度規劃，或決定哪些公司先投入 AI 時。觸發詞包含「AI readiness」、「AI opportunity scan」、「where should we deploy AI」、「AI across the portfolio」、「AI quick wins」或「which portcos are ready for AI」。
---

# 投資組合 AI 就緒度

## 工作流程

### 步驟 1：連接投資組合資料

先問使用者投資組合的資料放在哪裡。不要自己假設，把選項列出來：

- **MCP server**：data room、SharePoint、Google Drive，或已連接的投資組合營運資料庫
- **本機檔案**：磁碟上的資料夾路徑，裡面有季度簡報、財報、董事會資料包
- **檔案上傳**：把 PDF、PowerPoint 或 Excel 直接拖進對話裡

連接好之後，抓取投資組合（或其中一部分）的季度更新、董事會簡報和財報。每家公司都要擷取：產業、營收、各職能的人數、提到的技術堆疊，以及已經在進行的 AI/自動化專案。

如果使用者只給一家公司，還是照樣掃描，但跳過跨投資組合的排序。

如果從資料看不出來，一開始就要問：
- 每家公司剩餘的持有期間（距離出場 12 個月時，AI 的回收期就沒那麼重要了）
- 有沒有哪家 portco 已經部署過、而且有效的東西

### 步驟 2：逐家公司掃描

每家公司都回答三個關卡問題。三個都是 yes → **Go**。任何一個是 no → **Wait**，並註明什麼條件能解開。

1. **資料到位了嗎？**他們能不能為這個使用情境提供乾淨的輸入（客戶清單、發票資料流、合約庫），而不用先花 6 個月做資料專案？
2. **有負責人嗎？**管理團隊裡要有人真的會推動，而不是只會「支持」的贊助者。
3. **30 天內能試辦嗎？**一個團隊、一個流程、用現成的工具。如果答案是「首先我們得要……」開頭，那就不是 quick win。

接著找出前 2-3 個槓桿點。在成本結構和營運裡找這些模式：

**後勤（Back Office，通常最快能試辦）**
- 發票處理、AP/AR 對帳、費用分類
- 合約摘要：廠商合約、租約、客戶 MSA
- 月結：對帳、波動說明、給貸方的報告初稿

**營收 / 前線（Front Office）**
- RFP 和提案初稿，如果營收是以專案為主，這是很大的槓桿
- 業務通話摘要與 CRM 資料整理
- 客服工單分流與首次回覆草稿
- 組態式 / 複雜產品的報價

**營運（依產業而定）**
- SOP 與品質文件的產生
- 排程與派工（現場服務、物流）
- 程式碼產生與 review（軟體類 portco）

每個槓桿點用一行記錄：它取代什麼、每週節省的 FTE 小時（假設 30-50%，不要抓 100%），以及是買現成的還是需要輕量開發。

### 步驟 3：跨投資組合排序

把每家公司的每個槓桿點全部疊成一份清單。依以下排序：

1. **金額影響**：年化的 EBITDA 貢獻（成本節省 + 營收提升，扣掉工具成本）
2. **見效速度**：到第一個可衡量結果的月數
3. **機率**：針對資料品質、變革管理風險、管理團隊能力打折

同分時的判斷：優先處理剩餘持有期間不到 18 個月的機會，因為這些現在就要動，不然就不用動了。

輸出這份排序：

| 排名 | 公司 | 機會 | 預估 EBITDA ($) | 見效月數 | 關卡 | 第一步 |
|---|---|---|---|---|---|---|
| 1 | | | | | Go | |
| 2 | | | | | Go | |
| 3 | | | | | Wait — [blocker] | |

### 步驟 4：找出可複製的做法

投資組合裡槓桿最高的動作，是把一個成功的打法在多家公司跑一遍。要掃描的：

- **同產業、同職能**：兩家醫療服務的 portco 都在手動處理事前授權（prior-auth）？一次實作，兩邊部署。
- **同樣的工具、不同的公司**：如果某家 portco 已經有運作中的發票處理方案，就把其他所有 AP 量大於 $Xm 的 portco 標記為快速跟進者。
- **共同的廠商議價力**：三家 portco 買同一個工具，就是一場議價的談判。

列出每個可複製的做法，標明領頭公司（負責驗證）和跟進公司（負責複製）。

### 步驟 5：輸出

給 operating partner 的一頁，結構配合投資組合檢討：

1. **整個投資組合的前 5 名**：步驟 3 那張排序表，附上負責人和 30 天的第一步
2. **可複製的做法**：2-3 個可以一次打到多家公司的 playbook
3. **各公司的 Go / Wait**：每家一行；Wait 的要寫是什麼條件能解開
4. **我們不做的事**：看起來紙上很漂亮、但沒通過關卡的機會；這樣 operating partner 不用每季重新吵一次
5. **總 EBITDA 貢獻**：整個投資組合的 AI 機會總額，拆成第 1 年的 quick win 與第 2-3 年的規模化

## 重要注意事項

- **依金額排序，不要看興奮程度。**一個無聊的 AP 自動化，在營收 $40m 的公司省下 $400k，每次都贏過華麗的面向客戶的聊天機器人。
- **真正卡住的限制幾乎都是資料，不是模型。**如果公司產不出乾淨的客戶清單，第一個專案就不是 AI，而是資料清理。要直接講明白。
- **先用現成的。**對沒有工程深度的公司來說，客製開發又慢、又貴、又脆弱。優先選他們買得到、部署得起來的工具。
- **有沒有負責人才是真正的關卡。**沒有內部負責人的 quick win，90 天內就會死。如果管理團隊裡沒人想要，不管金額多大，都標成 Wait。
- **持有期間決定急迫性。**離出場還有 3 年的公司，負擔得起打基礎的資料專案。離出場 12 個月的公司，需要的是能在 CIM 的 LTM EBITDA 裡看得到的東西，不然就跳過。
- **失敗的試辦本身就是訊號。**如果管理層已經試過什麼但沒有持續，在提出同樣的東西之前，先查出原因。

---
name: ai-readiness
description: Scan the portfolio for the highest-leverage AI opportunities and rank where to deploy operating-partner time. Ingests quarterly updates and financials across multiple portfolio companies, identifies quick wins at each, and stacks them into a single ranked action list. Use during quarterly portfolio reviews, annual planning, or when deciding which companies get AI investment first. Triggers on "AI readiness", "AI opportunity scan", "where should we deploy AI", "AI across the portfolio", "AI quick wins", or "which portcos are ready for AI".
---

# Portfolio AI Readiness

## Workflow

### Step 1: Connect to Portfolio Data

First, ask the user where the portfolio materials live. Don't assume — offer the options:

- **MCP servers** — data room, SharePoint, Google Drive, or a portfolio-ops database if one is connected
- **Local files** — a folder path on disk with quarterly decks, financials, board packs
- **File uploads** — drag PDFs, PowerPoint, or Excel directly into the conversation

Once connected, pull quarterly updates, board decks, and financials for the portfolio (or a subset). For each company, extract: sector, revenue, headcount by function, tech stack mentioned, and any AI/automation initiatives already in flight.

If the user provides a single company, still run the scan but skip the cross-portfolio ranking.

Ask up front if not obvious from materials:
- Hold period remaining per company (AI payback matters less 12 months from exit)
- Whether any portco has already deployed something that worked

### Step 2: Per-Company Scan

For each company, answer three gate questions. All three yes → **Go**. Any no → **Wait** with a note on what unblocks it.

1. **Is the data there?** Can they produce a clean input for the use case — customer list, invoice feed, contract repository — without a 6-month data project first?
2. **Is there an owner?** Someone on the management team who will drive this, not a sponsor who will "support" it.
3. **Can we pilot in 30 days?** One team, one workflow, off-the-shelf tooling. If the answer starts with "first we'd need to...", it's not a quick win.

Then identify the top 2-3 leverage points. Look for these patterns in the cost structure and operations:

**Back Office (usually fastest to pilot)**
- Invoice processing, AP/AR matching, expense categorization
- Contract abstraction — vendor agreements, leases, customer MSAs
- Month-end close: reconciliations, flux commentary, lender reporting first drafts

**Revenue / Front Office**
- RFP and proposal first drafts — big lever if revenue is project-based
- Sales call summaries and CRM hygiene
- Customer support ticket triage and first-response drafting
- Quoting for configured / complex products

**Operations (sector-dependent)**
- SOP and quality documentation generation
- Scheduling and dispatch (field services, logistics)
- Code generation and review (software portcos)

For each leverage point, capture in one line: what it replaces, FTE-hours/week saved (assume 30-50%, not 100%), and whether it's buy-off-the-shelf or needs a light build.

### Step 3: Rank Across the Portfolio

Stack every leverage point from every company into one list. Rank by:

1. **Dollar impact** — annualized EBITDA contribution (cost out + revenue lift, net of tool cost)
2. **Speed to value** — months to first measurable result
3. **Probability** — discount for data quality, change management risk, management team capability

Tiebreaker: favor opportunities with <18 months of hold period remaining — those need to move now or not at all.

Output the stack:

| Rank | Company | Opportunity | Est. EBITDA ($) | Months to Value | Gate | First Step |
|---|---|---|---|---|---|---|
| 1 | | | | | Go | |
| 2 | | | | | Go | |
| 3 | | | | | Wait — [blocker] | |

### Step 4: Find the Replays

The highest-leverage move in a portfolio is running one successful play at multiple companies. Scan for:

- **Same sector, same function** — two healthcare services portcos with manual prior-auth? One implementation, two deployments.
- **Same tool, different company** — if one portco already has a working invoice-processing setup, flag every other portco with >$Xm in AP volume as a fast follower.
- **Shared vendor leverage** — three portcos buying the same tool is a pricing conversation.

List each replay with the lead company (who proves it) and follower companies (who copy it).

### Step 5: Output

One page for the operating partner, structured for a portfolio review:

1. **Top 5 across the portfolio** — the ranked table from Step 3, with owner and 30-day first step
2. **Replays** — 2-3 playbooks that hit multiple companies at once
3. **Go / Wait by company** — one line each; for Waits, what unblocks them
4. **What we're NOT doing** — the opportunities that looked good on paper but failed a gate; saves the operating partner from relitigating them every quarter
5. **Aggregate EBITDA contribution** — total portfolio-wide AI opportunity, split Year 1 quick wins vs. Years 2-3 scale

## Important Notes

- **Rank by dollars, not excitement.** A boring AP automation that saves $400k at a $40m revenue company beats a flashy customer-facing chatbot every time.
- **The binding constraint is almost always data, not models.** If a company can't produce a clean customer list, AI isn't the first project — a data cleanup is. Say so plainly.
- **Off-the-shelf first.** Custom builds are slow, expensive, and fragile for companies without engineering depth. Favor tools they can buy and deploy.
- **Ownership is the real gate.** A quick win with no internal owner dies in 90 days. If no one on the management team wants it, mark it Wait regardless of the dollar size.
- **Hold period drives urgency.** A company 3 years from exit can afford a foundational data project. A company 12 months out needs something that shows up in the LTM EBITDA for the CIM — or skip it.
- **Failed pilots are signal.** If management already tried something and it didn't stick, find out why before proposing the same thing again.