cx-pii-redaction-auditlisted

# Auditing PII redaction Support transcripts are among the densest personal data a company holds. In a single thread you can find a name, an email, a postal address, a phone number, a partial card number, a date of birth, an account balance, a reason for a payment, a health condition explaining a refund request, and a photograph of a document. That makes a redaction audit the gate before support data moves anywhere: to a vendor, into an analytics warehouse, into a model's context, or to a wider internal audience. **This audit does not make data safe. It measures how unsafe it still is**, so someone with the authority can decide. Do not present a clean result as clearance to share. ## Establish the purpose and the standard first Redaction is only meaningful against a stated purpose. "Enough for internal analytics" and "enough to send to a third-party model provider" are very different bars. Write down: - **Who will see the data**, and under what agreement. - **What the data will be used for**, and whether that use is compatible with why it was collected. This is a legal question in most jurisdictions, and it is not one you answer — flag it. - **Whether re-identification matters.** Removing names does not anonymise a dataset. A conversation is often uniquely identifying from its content alone: a specific transaction, an unusual complaint, a distinctive combination of dates and amounts. **Redaction is not anonymisation**, and claiming otherwise is the most conse