Get the whitepaper!
The Test Data Imperative: Towards SAP Autonomous Enterprise

Format · Documents

Mask PDFs, Word and PowerPoint files before they leave production

Statements, contracts, KYC packs, claim forms and board papers carry names, account numbers and addresses on every page. Synthesized finds them in PDF, DOCX and PPTX files, tokenizes, replaces or redacts each one, and checks the output against what it expected to find.

Get a 48-hour proofSample documents
direct-debit-letter.pdfMasked in your VPCIllustration
Before
Dear Maria Lopez, 22 Elm Road, Leeds LS6 2AB. IBAN GB33 BUKB 2020 1555 5555 55 will be debited on 1 October.
After
Dear Ava Okafor, 9 Mill Lane, York YO1 7HB. IBAN GB00 TEST 0000 0000 0000 00 will be debited on 1 October.
Found in this file
Treatment
Person name×1Consistent
Address×1Replace
IBAN×1Keep format
Debit date×1Keep
Policy: Customer data v3 Labelled-sample check passed

What's in these documents

NamesAddressesDates of birthIBANs and account numbersCard numbersPassport and ID numbersPhone numbers and emailsPolicy and claim numbersFree-text comments

Where they live

  • Document management systems and SharePoint exports
  • Loan, mortgage and claims files
  • KYC and onboarding archives
  • Contract repositories
  • Attachments stored next to database records

What you can do today

Engineer-ledMask documentsPDF, DOCX and PPTX, one format per batch. Choose per category whether to tokenize, replace, redact or retain.
Engineer-ledScanned pagesScans go through OCR first. Handwriting, tiny or stylised text and text over busy backgrounds can be misread, so check scanned batches with a labelled sample.
Engineer-ledSynthetic documentsFill document templates with synthetic values to get files with no real person behind them. See samples.
Engineer-ledZIP archivesWork when every file type inside has an adapter. A readable archive is not proof that everything in it was masked.
Engineer-ledSpreadsheets, web pages and imagesXLSX, CSV, HTML, plain text and calendar files, plus the text inside PNG and JPEG images and images embedded in Office files and PDFs, read with OCR and masked in place. Delivered in projects run with our engineers.

Know the limits

  • Masked PDFs come out as images, so they are not text-searchable afterwards.
  • Word and PowerPoint files with unsupported embedded content are refused rather than passed through half-masked.
  • One format per batch: run PDFs and Word files as separate batches.
  • A passed check is not proof that nothing was missed. Measure coverage on labelled samples.

Used for

Testing document AI, IDP and OCRDSAR and FOI redactionSharing and licensing dataKeeping data out of LLMsDev and test environments

Questions about document masking

Is the original text still inside the masked PDF?

No. Masked values are replaced in the file itself, so there is nothing underneath to copy out. That avoids the classic redaction failure where a black box sits on top of text that is still there. Output can be image-based, with no text layer, or searchable, with the replacement values in the text layer.

Can I mask a mix of PDFs and Word files?

Yes, as separate batches: each batch takes one format. Put batches in the same consistency group when the same people appear across them.

Do the documents leave my environment?

No. Extraction, detection and reconstruction run in the processing environment you configure, and the token-to-value mapping stays local.

The platform underneath, in production

Documents run on the same platform banks and insurers already trust with their test data.

Documents and text use the same policies, the same engine and the same deployment as these results.

Banking · test data80%less time to provision test data, with compliant data generated through the API inside CI/CD.Deutsche BankCase study
Insurance · security0security waivers across 14 priority applications: 28M+ rows secured, 100% referential integrity, test data 60% faster.Global specialty insurerCase study

Published results from database and synthetic data work on the Synthesized platform.

See it on your own documents

Bring a sample set. We'll show what was found, what was changed and what still needs checking.

Get a 48-hour proof