OpenAI B2B Signals
前沿優勢正開始不斷累積、持續放大。
今天，我們推出 B2B Signals，這是 OpenAI Signals 的企業版延伸內容，用於衡量 AI 如何在機構中普及。早期訊號很明確：前沿企業正拉大領先差距，不只是因為能取得 AI，而是因為能在各項工作中更深入地使用 AI。
B2B Signals 是一套定期衡量指標，基礎為針對企業 AI 使用情形進行經隱私保護的大規模分析。這套指標追蹤行為與模式，協助機構了解如何將智慧轉化為商業價值。
在前沿企業（即 AI 使用程度位於第 95 百分位的企業），每位員工使用更多 AI 智慧，也更密集採用進階工具，並將 AI 更深入融入工作流程。對部分企業而言，差距正開始累積放大，而差異也更加關乎使用深度。
重點摘要
- 前沿優勢開始出現累積效應：前沿企業如今每位員工使用的 AI 智慧是一般企業的 3.5 倍，高於一年前的 2 倍。
- 前沿企業更深入使用 AI，而不只是更頻繁使用：訊息量只能解釋前沿企業與一般企業之間 36% 的差距。前沿優勢大多來自更深入的使用方式。
- 智慧體工作流程正成為前沿採用程度的指標：差距在進階智慧體工具中最為顯著，前沿企業傳送的 Codex 訊息數量是一般企業的 16 倍。
- 企業可以透過組織變革縮小前沿差距：若要迎頭趕上，企業需要衡量使用深度、優先推動治理、投資於 AI 採用支援、擴大有效做法，並從對話式協助轉向將工作委派給智慧體。
深度
前沿優勢開始出現累積效應，而最深入運用 AI 的企業正持續拉大領先差距
對企業而言，席位部署只是起點。更值得觀察的是，員工是否已開始運用 AI 處理更深入、更複雜的工作。此圖表比較前沿企業（定義為第 95 百分位）與一般企業（定義為第 50 百分位）每位工作者產生的 Token 數量。
Token 並非衡量商業價值的完美指標。簡短的回應可能非常有價值，而冗長的回應也可能價值不高。但 Token 使用量有助於衡量員工交給 AI 處理多少工作，因此能作為 AI 使用深度，以及員工對 AI 智慧需求程度的實用參考指標。
前沿企業每位工作者對 AI 智慧的需求量，是一般企業的 3.5 倍。這項差距已從 2025 年 4 月的 2 倍進一步擴大，顯示 AI 應用最深入的企業正持續拉大領先差距，也更有能力將新的 AI 能力轉化為更深入、更複雜的工作。
前沿優勢大多來自更深入的使用，而非更高的訊息量
前沿企業每位員工所需的 AI 智能大幅高於一般企業，但大部分差距不能單以訊息量來解釋。此圖表拆解了前沿企業 3.5 倍的優勢，結果發現，即使一般企業也像前沿企業一樣頻繁傳送訊息，也只能縮小其中 36% 的差距。
其餘差距則來自更深入的使用方式。前沿工作者會要求 AI 承擔更複雜的工作、為模型提供更豐富的脈絡，並產生更具實質內容的輸出。
廣度
The frontier advantage is largest in advanced agentic tools, demonstrated by a 16x Codex usage gap
The frontier advantage is largest for tools that support more advanced workflows. Codex shows the largest gap, with the frontier firm sending 16x as many messages per worker as the typical firm. ChatGPT Agent, Apps in ChatGPT, Deep Research, and GPTs also show relatively large gaps, suggesting that the frontier is better at leveraging tools that help workers code, delegate multi-step tasks, apply company context, and conduct more complex research.
By contrast, more general-purpose and accessible tools such as User Upload, Search, and Data Analysis show a smaller frontier advantage. These tools are easier for most firms to use because they extend familiar workflows. The frontier advantage is most pronounced in advanced and agentic tools, where adoption requires more expertise, connections to workplace knowledge and tools, and greater comfort delegating work to AI.
The largest frontier advantage is in education and learning
The frontier advantage is largest for education and learning tasks, where the frontier firm sends 7x as many messages per worker as the typical firm. At the frontier, firms use AI to help employees build skills and learn new topics. They also use AI to improve their understanding of AI itself, including what it can do, how to use it well, and where it can fit into existing workflows. The size of the gap suggests that the typical firm may underutilize AI as a tool for workforce learning and development.
Coding also shows a large frontier advantage, with the frontier firm sending 4x as many messages per worker as the typical firm. This is consistent with the broader gap in advanced and agentic tool use. How-to guidance and writing and communication have the smallest frontier gaps, likely because these tasks are more accessible and familiar uses of AI.
要縮小能力落差，關鍵在於支援 AI 採用，而不只是開放 AI 使用。OpenAI 的 企業資源 和 OpenAI Academy 提供實用指南、訓練教材和部署資源，協助團隊安心採用 AI。
AI use is broadest in writing but function-specific trends are emerging
Writing and communication remain the most common uses of ChatGPT. However, usage patterns vary by function and are often tied to each function’s core responsibilities. 60% of IT & Security messages are concentrated in how-to and procedural guidance, almost half of Software Development and Data Science & Engineering messages are related to coding, and a tenth of Finance messages are related to analysis and calculation.
These patterns are consistent with broader evidence that frontier models are improving on economically valuable workplace tasks. GDPval, an evaluation of real-world knowledge work across 44 occupations, measures performance on tasks that produce practical work outputs such as documents, spreadsheets, slides, diagrams, and multimedia. As AI becomes more capable, enterprise usage appears to be extending toward tasks that are more closely tied to each function’s core work.
依業務情境劃分的任務類型
| 業務情境 | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ChatGPT 任務 | ||||||||||||
| 寫作與溝通 | ||||||||||||
| 操作與流程指南 | ||||||||||||
| 資訊 | ||||||||||||
| 分析與計算 | ||||||||||||
| 建議 | ||||||||||||
| 創意媒體 | ||||||||||||
| 商業 | ||||||||||||
| 寫程式 | ||||||||||||
| 教育與學習 | ||||||||||||
觸及範圍
Industry leadership is not one-dimensional: different sectors lead across ChatGPT, Codex, and the API
There is no single AI adoption leaderboard. Industry rankings vary depending on the measure used. Professional, Scientific, and Technical Services ranks first in both Codex adoption and API intensity, indicating relatively advanced use in developer and product-integrated workflows. Finance and Insurance leads in ChatGPT adoption due to large-scale deployments, while Educational Services has the highest message intensity, suggesting deeper per-person usage. Retail Trade and Health Care rank highly in API intensity, despite lower rankings on other measures.
These differences suggest that industry leadership is not one-dimensional. Some sectors appear to be adopting AI through technical and developer workflows, while others are scaling through broad ChatGPT adoption or more intensive end-user usage.
依 AI 採用指標劃分的產業排名
| 產業 | ||||
|---|---|---|---|---|
| 金融與保險 | 1+1 | 10-4 | 30 | 60 |
| 資訊 | 2-1 | 20 | 20 | 4-1 |
| 專業、科學及技術服務 | 30 | 10 | 10 | 10 |
| 藝術、娛樂與休閒 | 40 | 4-1 | 50 | 3+1 |
| 公用事業 | 50 | 80 | 90 | 90 |
| 營造 | 6-1 | 50 | 10-1 | 10-1 |
| 房地產與租賃 | 7-1 | 7+1 | 11-1 | 80 |
| 製造 | 8-1 | 3+1 | 40 | 70 |
| 醫療保健與社會援助 | 90 | 90 | 6+1 | 50 |
| 零售貿易 | 10-2 | 11-1 | 7-1 | 20 |
| 公共行政 | 11-1 | 6+1 | 80 | 11-1 |
企業正將 API 應用導入正式環境工作流程與客戶使用的應用程式
企業越來越常使用 API，將模型直接整合到產品、服務及內部系統中。正式環境中的常見使用案例包括應用程式內助理、程式碼編寫與開發工具、客戶支援、研究工作流程，以及工作流程自動化。
這些部署案例顯示，企業 AI 正逐漸走出實驗階段，開始導入可重複執行、且能帶來可衡量營運效益的工作流程。從各個客戶案例可見，企業正運用 OpenAI 模型加速知識工作、提升工程產能，並為客戶與員工打造 AI 驅動的體驗。
各產業最熱門的 API 使用案例
專業服務
知識庫助理與搜尋（例如 Q&A 工具、研究助理、內部知識庫助理）
客戶與銷售支援（例如客戶支援、語音與對話智慧體、銷售協助）
資料分析、摘要與擷取（例如公司資料分析、市場情報、交易標記與對帳）
寫程式與開發工具（例如：模型評估工具、程式設計助理、工作流程自動化工具）
金融和保險
資料分析、摘要與擷取（例如：資料擷取、收據與費用分析、投資研究）
文件與工作流程生成（例如自動化支出管理、研究摘要生成、工作流程最佳化）
知識助理與搜尋（例如投資策略助理、政策搜尋、特定職務助理。）
客戶與服務支援（例如客戶支援語音與對話智慧體、個人銀行助理、情緒分類）
資訊
寫程式與開發工具（例如程式碼編寫助理、軟體測試工具、網頁自動化工具）
知識助理與搜尋（例如產品內助理、內部搜尋工具、文件助理）
客戶與服務支援（例如客戶支援語音和對話智慧體、多通路客戶服務自動化）
內容、媒體與設計生成（例如品牌素材生成、行銷工具）
Cisco 使用 Codex，在大型企業工程組織中加速處理複雜的軟體工程工作。在正式環境工作流程中，Codex 協助將建置時間縮短約 20%，每月節省超過 1,500 小時的工程工時，並將缺陷修復吞吐量提升 10 至 15 倍。正如 Cisco 團隊所說，當他們把 Codex 視為「團隊的一分子」時，成效最為顯著。
Rakuten 在工程營運與軟體交付流程中部署 Codex，將平均復原時間縮短約 50%，並讓團隊能以兩倍速度解決正式環境問題。Rakuten 也使用 Codex 進行符合內部標準的自動化程式碼審查與弱點檢查，協助加快版本發布，同時不犧牲安全性。在複雜專案中，Codex 可以將部分需求轉化為可運作的全端實作，將時程從數季縮短至數週。
Balyasny Asset Management 運用 OpenAI，加速大型專業知識型機構中的投資研究流程。Balyasny 自有 AI 研究平台已由約 95% 的投資團隊採用，並協助將研究流程從數天縮短至數小時。例如，中央銀行演講分析流程過去需要兩天，如今約 30 分鐘即可完成，協助分析師更快速整合申報文件、逐字稿、研究報告與市場資料中的資訊。
如需查看更多案例，請造訪我們的客戶案例頁面。
What organizations can do to reach the frontier
OpenAI works with enterprises across industries, functions, and stages of AI maturity, giving us visibility into how adoption develops from experimentation to production. Across these deployments, the firms making the most progress tend to focus less on access alone and more on the organizational systems needed to use AI deeply: measurement, governance, enablement, scaling impact, and agentic deployment.
Five practices stand out as practical steps any organization can start taking today to deepen AI adoption.
- Measure depth of use in addition to access.
The relevant signal is not only how many employees have AI accounts, but whether teams are using AI more substantively over time. Organizations should track whether AI use is becoming more frequent, more complex, and more closely tied to valuable workflows. - Build governance that makes agentic AI deployable.
Leading firms are not avoiding governance. They are using it to make agentic AI more deployable. Firms need clear rules for where agents can operate, what information they can use, when they should advise rather than act, and how humans review higher-risk decisions. Frontier firms are defining these standards as part of the deployment process, so governance becomes a way to expand adoption safely rather than slow it down. - Treat enablement as core infrastructure, not a side project.
As AI capabilities improve, both workers and organizations need systems that help them keep pace. Frontier firms do not treat enablement as a one-time training push. They build continuous learning into deployment through role-specific training, use-case workshops, hackathons, internal champion networks, dedicated experimentation time, and shared repositories of workflows, best practices, and skills. - Identify your frontier teams and scale their impact.
In many organizations, the most advanced usage is concentrated in a small number of teams. Those teams can reveal which workflows, habits, and operating models are working. Leaders should identify these teams, understand and scale the conditions behind their success, and help them share insights and examples of deeper AI use with the rest of the firm. - Move beyond chat to delegating work.
Enterprise AI is shifting from chat assistants to work that can be delegated to agents. Software engineering illustrates this trend, but delegated work is spreading across functions. With Codex, engineers can hand off a defined task, give the agent the context it needs, let it work across files, codebases, and tools, then review the result and refine the workflow with feedback. Frontier firms are encouraging workers to delegate tasks to AI rather than simply using AI as a static assistant.
本報告中的所有分析，均以去識別化並彙總的企業使用資料為依據。訊息內容已使用自動化系統進行分類，且在此分析過程中，沒有任何 OpenAI 員工審查個別 Enterprise、Business 或 API 客戶資料。
If you’d like to explore the full findings or learn how to bring AI into your organization responsibly, we’d love to connect.
