Skip to content

Redact account and customer data before it is shared.

Keep statements, applications, and reports clean without a cloud service.

Where sensitive data leaks out

Financial teams handle account numbers, IBANs, and customer identities that attract both fraud and regulatory scrutiny. Pasting these into AI assistants or converting files online puts sensitive data outside the institution.

PCI DSS tells you how much of a card number may survive

Financial data attracts unusually specific rules, and the specificity is the useful part. PCI DSS version 4 draws a line most teams blur: masking and truncation are different obligations with different permanence. Requirement 3.4.1 governs display, where at most the first six and last four digits of a primary account number may be visible. Requirement 3.5.1 governs storage, where a stored number must be rendered unreadable by strong encryption, truncation, tokenisation or one-way hashing. Masking is temporary and cosmetic. Truncation is permanent and destroys the middle digits for good. Choosing the wrong one is how a compliant-looking export turns out to hold full card numbers underneath.

Display: first six and last four, at most

The BIN and the last four are enough to reconcile a transaction with a customer on the phone. Anything more is over-exposure. The MASK operator preserves exactly the segments you configure, so a statement can be shared for reconciliation without carrying the full number.

Storage: unreadable, by one of four methods

Encryption, truncation, tokenisation or one-way hashing. Omit covers three directly: HASH for one-way, MASK for truncation, and the encrypted vault for reversible tokenisation where the value must stay recoverable under control.

Reversibility is a deliberate choice, not a default

Hashing a card number closes the door permanently. Vault-backed tokenisation keeps it openable by authorised staff. Both are legitimate; the mistake is picking one by accident. Policy is set per entity type so the decision is made once and applied consistently.

The documents this actually applies to

Statements and transaction exports

CREDIT_CARD, IBAN, ACCOUNT_NUMBER, SORT_CODE, PERSON

Routinely exported to spreadsheets for reconciliation, at which point they leave the controls of the core banking system. Omit handles CSV and Excel at the cell level, so account columns can be masked while amounts and dates stay intact and the analysis still works.

Know your customer and onboarding files

PERSON, DATE_OF_BIRTH, national ID numbers, ADDRESS, scanned identity documents

Dense with identifiers by design, since proving identity is the purpose. Most of the content is scanned images, which is where OCR matters: an identity document redacted only in the text layer is not redacted at all.

Loan, mortgage and credit files

PERSON, ACCOUNT_NUMBER, IBAN, employer and salary details

These files travel between broker, lender and underwriter, and every hop is an opportunity to over-share. Redacting to the minimum the recipient needs is easier than negotiating what each party may retain.

Advice notes and client correspondence

PERSON, ACCOUNT_NUMBER, holdings and positions

Identifiers sit in running prose rather than fields, and advisers paste extracts into general purpose AI tools to draft summaries. Protect redacts on paste, so the model receives the substance without the account.

Masking a statement line to the PCI display rule

Before

Payment from Helena Vasquez, card 4539 8812 7734 9061, IBAN GB29 NWBK 6016 1331 9268 19, settled 14 June 2026 for 2,480.00.

After

Payment from [PERSON_1], card 453988XXXXXX9061, IBAN [IBAN_1], settled 14 June 2026 for 2,480.00.

The BIN and last four survive on the card, which is what the display rule permits and what a reconciliation actually needs. The amount and settlement date are untouched, because neither is an identifier and removing them would make the record useless. The IBAN is tokenised rather than masked, because unlike a card number there is no partial form of an IBAN that is both useful and safe.

Showing which method was applied to which field

A PCI assessment turns on whether the right treatment was applied in the right place, not on whether something was redacted. The per-run report records each detection, its type and the operator used, so an assessor can see that stored numbers were hashed or truncated while displayed numbers were masked. Because the report is written during the run, it is contemporaneous evidence rather than a reviewer's later recollection of intent.

What records teams ask first

Can it keep the BIN and last four rather than removing the whole card number?

Yes. The MASK operator preserves the segments you configure, so first six and last four is a policy setting rather than a manual edit. That is the maximum PCI DSS permits on display, and it is enough to reconcile a payment with a customer.

Does it work on spreadsheets without breaking them?

Omit rewrites cell values in CSV and Excel rather than treating the file as flat text, so structure survives. Amounts, dates and reference columns are left alone unless policy targets them, which is what keeps the export usable after redaction.

Hashing or vault tokenisation for stored numbers?

Hash when the original is genuinely never needed again, because irreversibility is the strongest option. Use the encrypted vault when an authorised person must resolve the token later, and treat the result as still sensitive: a recoverable mapping means the data has been pseudonymised rather than removed.

What about scanned identity documents?

Bundled OCR runs locally and detections are mapped back to coordinates on the image, so the redaction is applied to the picture rather than only to an extracted text layer. This is the common failure in KYC workflows.

Does client financial data leave our network?

No. Detection and redaction run on the machine holding the file, with no upload and no service dependency. For a regulated firm that removes an entire vendor assessment, because there is no vendor endpoint receiving client data to assess.

Can different desks have different policies?

Yes. Policy is configured per entity type and administrators can set a floor local users cannot widen. A reconciliation desk can be permitted masked card numbers while a marketing analytics export is forced to hash them.

Further reading

  • How to Redact PII From CSV, Excel, and JSON

    Omit redacts PII inside structured data files, not just documents: CSV and Excel rewrite the cell, JSON rewrites the value in place, and in every case the file's shape survives. Here is exactly what changes and what does not, format by format.

  • What Is PII? Definition, Examples, PII vs PHI

    PII means information that identifies a person, but four different laws draw that line in four different places, and the same email address can be in scope under one and arguable under another. Here is each definition, what counts as an example of each, and where PII ends and PHI begins.

  • Redaction vs Anonymization vs Pseudonymization

    Four words that get used interchangeably and mean four different things, with four different legal consequences. Here is what each one actually does to your data, which regulation defines it, and which one your tool is really performing.

  • How to Unredact a PDF, and How to Prevent It

    Most recovered redactions come back the same trivial way, because a black rectangle is a drawing and the text is still underneath it. Here is why it happens, how to tell which kind you are holding, and what actually removes a value.

On-device processing keeps customer financial data inside the institution, supporting GDPR and internal data-handling controls, with fail-closed behavior when anything is uncertain.

Keep customer data inside the bank.