Excel + AI: A 5-Shot Guide to Using Personal Data Safely
When using AI features such as Copilot in Excel, exposing personal data as-is can create significant risk. This article provides a practical five-shot guide to handling personal data safely.
Summary
| Step | Key Point |
|---|---|
| Recognize the risk | Do not enter raw personal data directly into AI |
| Data minimization | Use anonymous IDs instead of names and phone numbers |
| Pseudonymization | Mask data with functions such as LEFT/RIGHT and REPLACE |
| Protection | File encryption, access control, and log management |
| Advanced techniques | Differential Privacy and Federated Learning |
Shot 1 — Recognize the Risk
If you directly enter data containing raw personal information into Copilot, such as =COPILOT("Summarize return rates by customer name", Orders[#All]), personally identifiable information may be exposed unintentionally.
Shot 2 — Data Minimization
Only **anonymized data** should be used as AI input.
// Before
Customer_Name: "Hong Gildong"
// After
Customer_ID: 10023Also revise prompts to use wording such as “return rate by customer ID.”
Shot 3 — Pseudonymization and Masking
Use Excel functions to mask personal information:
=LEFT(A2,1) & "OO" // Name masking (Hong Gildong → HOO)
=REPLACE(B2,4,4,"****") // Mask the last digits of a phone numberYou can apply the same process in Power Query using Replace Values or Add Column ▶ Transform.
Shot 4 — Encryption and Access Control
- Encrypt Excel files (File ▶ Info ▶ Protect Workbook)
- Set OneDrive/SharePoint permissions (read-only/restricted editing)
- Track who opened files by recording access logs
Shot 5 — Privacy-Enhancing Techniques
At the organizational level, apply Privacy-Preserving AI to AI training and analysis:
- Differential Privacy: Prevents individual tracking by adding noise to data
- Federated Learning: Keeps data local and shares only model updates