From pseudonymization to re-identification in 6 steps
Before a document can be safely submitted to an AI tool, it must be made "AI-ready": personal data is replaced with consistent pseudonyms, the AI processes the pseudonymized version, and then the original data is restored. This workflow comprises six clearly defined steps.
This article describes each step in detail — from document import to re-identification — and explains what matters at each stage.
PDF, Word, email — over 70 formats are supported
In the first step, the documents to be processed are imported into Docuflair Mask. The software supports over 70 file formats:
Documents can be imported individually or in batches. For scanned documents, OCR text recognition is performed automatically to make the text machine-readable.
Important: All documents remain on-premises. Import is performed locally — no document is transferred to external servers for pseudonymization.
9 categories of personal data are identified automatically
The software analyses the document content and detects personal data automatically. 9 categories are distinguished:
| Category | Detection method | Examples |
|---|---|---|
| Names | Dictionary + context analysis | John Smith, Dr Miller |
| Email addresses | Pattern recognition | john@company.co.uk |
| Addresses | Context-based | 123 Main Street, London EC1A 1BB |
| Phone numbers | Pattern recognition | +44 20 7946 0958 |
| IBAN numbers | Check digit validation | GB29 NWBK 6016 1331 9268 19 |
| Company names | Dictionary + context analysis | ABC Ltd, XYZ plc |
| Dates | Pattern recognition | 15/03/2026, 2026-03-15 |
| Tax numbers | Country-specific formats | VAT ID, tax reference |
| National insurance numbers | Country-specific formats | NI number, SSN |
Detection combines rule-based pattern recognition with dictionary matching and context analysis. Dictionaries can be individually maintained — for instance by importing from Active Directory or CSV files.
"John Smith" becomes "Person_A" in ALL documents
After detection, personal data is replaced with consistent pseudonyms. The core principle: same person = same pseudonym — across all documents and all processing runs.
If John Smith appears in 50 documents, he becomes Person_A everywhere. If his colleague Jane Brown appears in 30 documents, she becomes Person_B everywhere. This preserves relationships:
The AI recognises that the same person is involved throughout — without knowing their identity.
Consistency applies not only within a single processing run but also across batches. If John Smith was pseudonymized as Person_A in batch 1, he will also receive the pseudonym Person_A in batches 2, 3 and all subsequent runs. The replacement table is continuously extended.
The AI-ready document is prepared for handover to external tools
After pseudonymization, the document is exported. It contains exclusively pseudonyms — no real personal data. This AI-ready document can be safely submitted to external AI tools:
Since no personal data is transferred, the GDPR risk is minimised — regardless of which AI tool is used and where its servers are located.
The AI delivers a result based on the pseudonymized data
The AI tool processes the pseudonymized document and delivers a result. This result also contains only pseudonyms:
AI summary (pseudonymized): "The contract between Person_A and Company_A governs the delivery of Product_A. Company_A commits to delivery by Date_B. In case of delay, Person_A is entitled to a contractual penalty of Amount_B per day."
The summary is substantively complete and correct — only the personal data is pseudonymized. In the next step, the pseudonyms are replaced with the original data.
Pseudonyms are replaced with original data via the replacement table
In the final step, the pseudonyms in the AI result are replaced with the original data. Docuflair Mask uses the encrypted replacement table to turn Person_A back into John Smith, Company_A back into ABC Ltd and Address_A back into 123 Main Street.
Finished result: "The contract between John Smith and ABC Ltd governs the delivery of Product X. ABC Ltd commits to delivery by 30 April 2026. In case of delay, John Smith is entitled to a contractual penalty of EUR 500 per day."
See in a 15-minute demo how Docuflair Mask automates the entire workflow from pseudonymization to re-identification. On-premises and GDPR-compliant.
Answers to the most important questions about the pseudonymization workflow
Docuflair Mask supports over 70 file formats including PDF, Word, Excel, PowerPoint, scanned documents (TIFF, JPEG, PNG) and email formats (MSG, EML). The integrated OCR text recognition enables pseudonymization of scanned and image-based documents as well.
Docuflair Mask automatically detects 9 categories of personal data: names, email addresses, addresses, phone numbers, IBAN numbers, company names, dates, tax numbers and national insurance numbers. The categories are fully configurable and extensible.
Replacement tables are stored encrypted and kept separately from the pseudonymized document. Only authorised users have access. All access is logged in the audit trail. The table can additionally be exported and secured in an external vault.
Cross-batch consistency means that the same person receives the same pseudonym across multiple processing runs and document sets. If John Smith is pseudonymized as Person_A in batch 1, he will also receive the pseudonym Person_A in batches 2, 3 and all subsequent runs.