Practical guide
How to anonymise an Excel file with tax numbers before using it with AI
By Hugo Ribeiro · Updated · 6 min read
[ENTIDADE_1]), leave amounts and account codes untouched, and keep the lookup table on your own computer. Deleting columns or using Find and Replace is not enough: it misses numbers stored as numbers, hidden sheets, comments and the file name itself.A client list, an aged debtors report, a payroll listing: almost every spreadsheet in a practice holds tax numbers and names. To ask an AI for an analysis, none of that is needed. This guide shows how to take it out without breaking the sheet.
What must go and what can stay
- Goes: tax and VAT numbers, National Insurance or PPS numbers, names of people and companies, IBANs and account numbers, addresses and postcodes, emails, phone numbers, document numbers that lead to the client.
- Stays: amounts, dates, account codes, rates, quantities and the structure of the sheet. That is what the AI works on.
A company's registration or VAT number is not personal data, but it is covered by your duty of confidentiality if it reveals who the client is. In practice, remove both.
Step by step
- Work on a copy. Never on the original.
- Replace, do not delete. Each identifier becomes a stable code:
[DOC_ID_1],[ENTIDADE_1],[IBAN_1]. If you delete, the AI can no longer tell one client from another; if you replace, per-client analysis still works. - Always use the same code for the same party, across every sheet and every file in the engagement. "Smith & Sons Ltd" and "SMITH AND SONS LIMITED" are one party and must get one code.
- Keep the lookup table on your computer. It is what lets you put the real data back into the answer.
- Review before sending (list below).
- Ask the AI for the analysis and, at the end, restore the names and numbers in the answer.
Where data usually leaks
- Tax numbers stored as numbers, with different formats in the same column.
- Free text: transaction descriptions ("Transfer John Smith"), notes, narratives.
- Hidden sheets and hidden rows or columns.
- Comments and notes on cells.
- Sheet names ("Smith 2026") and print headers and footers.
- The file name and document properties (author, company).
- Pivot tables, which keep a copy of the source data even after you delete it from the sheet.
Check the last three by hand in Excel, under File > Info > Inspect Document, whatever tool you use.
Doing this with Velado
Velado does steps 2 to 4 and step 6 for you, on your computer. It opens the workbook, detects tax numbers (checking the check digit where the format has one, even when stored as numbers), names, IBANs and addresses, shows you what it is about to mask and exports a workbook with codes, with amounts and formulas in place. The lookup table stays in a local vault per client, and one click swaps the codes back in the AI's answer.
The review is still yours: no automatic detection is perfect, and the review screen is there so you can add what is missing by hand, such as an unusual trading name.
You can try it in the browser with a file of your own; nothing is sent to any server. For the full flow with a trial balance, see the trial balance guide.
FAQ
Can I just delete the tax number column?
It is not enough. The client's name, IBAN, address or document number identify them just as well, and they usually appear in other columns, in transaction descriptions and on other sheets of the same workbook.
Why does Find and Replace fail with tax numbers?
Because Excel stores many of them as numbers: it drops leading zeros, shows thousands separators or merges them into text ("VAT 123456789"). An exact text search does not catch every form, and one missed cell is enough to identify the client.
Is this anonymisation or pseudonymisation?
It is pseudonymisation: because you keep the table that leads back to the real data, the file is still personal data for you. For the AI provider, which does not have the table, the client can no longer be identified. Our guide on pseudonymisation and anonymisation explains the difference.
Do formulas and totals survive?
Yes, as long as only identifiers are replaced. Sums, percentages and cell references do not depend on names or tax numbers. Lookups by name (VLOOKUP) can break, which is why the same name must always become the same code.
Try it with your own document
Velado runs in your browser, on your computer: nothing is sent to any server.