Governing AI use管控 AI 的使用
Without an approved tool, staff paste internal data into public chatbots.
沒有獲批准的工具，員工就會把內部數據貼到公共聊天機械人。
Banking銀行 Insurance保險
Use AI on your data.用 AI 查數據，Safely.安全可控。
Staff ask in Microsoft Teams and get answers in seconds, only from data they are allowed to see.
員工在 Microsoft Teams 提問，幾秒得到答案，只限自己有權查看的數據。
Interactive recreation of SafeFlow in Microsoft Teams · Fictional dataSafeFlow 在 Microsoft Teams 中的互動示範 · 虛構數據
The challenge面對的挑戰
The question is no longer whether to use AI.問題不再是用不用 AI，It is how to use it safely.而是如何安全地用。
Managing permissions管理存取權限
AI must follow the same entitlements as your core systems: by role, team and client book.
AI 必須遵守與核心系統相同的權限：按角色、團隊和客戶組合。
Getting data fast快速取得數據
An MI report or a data request takes days. Decisions cannot wait.
一份報表或數據申請要等好幾天，但決策等不了。
How SafeFlow governs AISafeFlow 如何管控 AI
Fast answers, inside your rules.答案要快，規則不變。
Every question passes the same three checks before anyone sees an answer.
每個提問在給出答案之前，都要經過同樣的三重把關。
Employee員工
Asks in Microsoft Teams在 Microsoft Teams 提問
SafeFlow
Access check · Read-only query · Logged權限檢查 · 唯讀查詢 · 全程記錄
Your data你的數據
Databases, Excel, SharePoint, OneDrive數據庫、Excel、SharePoint、OneDrive
Access checked every time每次都檢查權限
Checked by role before every answer. Staff only see what they are allowed to.
每個答案前都按角色檢查權限，員工只看到有權看的數據。
Read-only by design設計上只讀不改
Every query is read-only. The AI can look data up, never change it.
查詢全部唯讀。AI 只能查，不能改。
A full audit trail完整審計記錄
Every question is logged: who asked what, and when.
每個提問都有記錄：誰、何時、問了甚麼。
Built by data protection experts. Our senior engineers previously built enterprise data protection products at a leading company in the field, where they earned 30+ patents in China and the US. 由數據保護專家打造。我們的資深工程師曾在業內領先的數據保護公司開發企業級產品，並在任職期間於中國和美國取得 30 多項專利。
Works in the chat tools you already approve在機構已批准的通訊工具中使用
Microsoft Teams
Slack
Google Chat
Same SafeFlow同一個 SafeFlow
Different priority.不同的重點。
For SMEs中小企業
Low cost, fast start低成本，快速開始
Build an AI tool for business data in weeks, inside WhatsApp, without a data team or a big project.
幾星期內在 WhatsApp 裏建立查數據的 AI 工具，不需要數據團隊，也不用大型項目。
For banks and insurers銀行及保險機構
Safe, governed adoption安全、可管控地採用 AI
Give staff AI on internal data in Teams, under the access rules, oversight and audit trail a regulated business needs.
在 Teams 讓員工用 AI 查內部數據，同時保持受監管機構所需的存取規則、監督和審計記錄。
Scenarios應用情境
One question away, for every team.每個團隊，一問就知。
Each person gets the answer their role allows, and nothing more.
每個人只會得到自己角色允許的答案，不多也不少。
Retail banking零售銀行
Branch and mortgage operations分行及按揭營運
Which branches have the most mortgage applications pending over 10 days?哪些分行待處理超過 10 天的按揭申請最多？
Two branches hold most of the backlog, and most of those cases are waiting on valuation reports.
兩間分行佔了大部分積壓個案，其中多數正在等待估價報告。
Applicant names and income: the case officer and their manager only.申請人姓名及收入：只限負責職員及其主管。
Wealth management財富管理
Relationship managers客戶經理
Which of my clients have time deposits maturing this month?我的客戶之中，哪些定期存款在本月到期？
Twelve clients have deposits maturing, and three of them have no follow-up meeting booked.
有十二位客戶的存款將到期，其中三位尚未安排跟進會面。
Client portfolios: each RM sees only their own book.客戶組合：每位客戶經理只看到自己負責的客戶。
Insurance claims保險理賠
Claims operations理賠營運
Which claims have been open for more than 30 days, by product line?按產品線劃分，哪些理賠個案已超過 30 天未結？
Medical claims make up most of the backlog, mostly waiting on documents from the claimant.
醫療理賠佔積壓個案的大部分，主要在等待申索人補交文件。
Medical details: the assigned claims handler only.醫療資料：只限負責的理賠人員。
Insurance distribution保險分銷
Agency and bancassurance代理及銀行保險
Which teams' first-year persistency fell this quarter?本季哪些團隊的首年保單繼續率下跌？
Two teams are below target, concentrated in policies sold through one campaign.
有兩個團隊低於目標，主要集中在同一推廣活動售出的保單。
Commission: each agent sees only their own.佣金：每位代理只看到自己的。
Finance財務
Management reporting管理報告
How did cost-to-income change by business unit this quarter?本季各業務單位的成本收入比有何變化？
One unit's ratio rose after a one-off system cost; the others held steady.
其中一個單位因一次性系統開支而上升，其餘單位保持平穩。
Unit-level P&L: finance and senior management only.各單位損益：只限財務及高級管理層。
Your team?你的團隊？
Bring one question your team waits days to answer. We will show you the answer, and who gets to see it.
帶一個團隊要等好幾天才有答案的問題來，我們會示範答案，以及誰可以看到它。
Book a briefing預約簡介Sample scenarios with fictional data.示例情境，所用數據均為虛構。
FAQ常見問題
Questions from risk and IT.風險及 IT 團隊常問的問題。
Can staff see data they are not entitled to?員工會看到自己無權查看的數據嗎？
No. Access is set by role and checked before every answer, so each person only gets answers from data they are allowed to see.
不會。存取權限按角色設定，並在每個答案給出前檢查，每個人只會得到自己有權查看的數據。
Can the AI change our data?AI 會更改我們的數據嗎？
No. Every question runs as a read-only query. SafeFlow can look data up but cannot write, update or delete it.
不會。每個提問都以唯讀方式查詢，SafeFlow 可以查閱數據，但不能寫入、修改或刪除。
Can we review what was asked?我們可以查閱提問記錄嗎？
Yes. Every question is logged, giving risk, compliance and internal audit a record of who asked what.
可以。每個提問都有記錄，讓風險、合規及內部審計團隊知道誰問了甚麼。
Can SafeFlow run in our own environment?SafeFlow 可以部署在我們自己的環境嗎？
Yes. SafeFlow ships as a self-contained deployment package that installs on your own servers. Each organisation runs its own separate instance, with its own storage and credentials.
可以。SafeFlow 提供獨立的部署套件，可安裝在你自己的伺服器上。每個機構都運行各自獨立的實例，擁有自己的儲存和憑證。
Which AI model is used, and is our data sent to it?使用哪個 AI 模型？我們的數據會傳送給它嗎？
You choose the model endpoint, for example Azure OpenAI, so it can follow your existing AI and vendor policies. The model never connects to your systems. To write an answer it sees the question and the results of queries the asker is allowed to run; data outside their access is never queried.
模型端點由你選擇，例如 Azure OpenAI，可配合機構現有的 AI 及供應商政策。模型不會直接連接你的系統。撰寫答案時，它只會看到提問以及提問者有權執行的查詢結果；權限以外的數據從不會被查詢。
See it on your use case.用你的情境親身體驗。
A 30-minute briefing for business, risk or IT teams: how SafeFlow answers, and how it stays inside your rules.
為業務、風險或 IT 團隊而設的 30 分鐘簡介：SafeFlow 如何給出答案，又如何守住你的規則。