De-identification usually proceeds in layers. Direct identifiers are removed or replaced first: names, email addresses, phone numbers, account and government numbers. Join keys that must be kept are replaced with salted hashes. Indirect identifiers are then reduced by generalizing values, for example a full date to a year, or by removing outlying records. Free text needs both pattern matching and human review of samples, because identifying details hide in sentences.
The term is used most in United States law. HIPAA defines two methods for health information, Safe Harbor and Expert Determination. The CCPA defines deidentified information and attaches duties to the business that holds it. Data that meets a US de-identification standard is not automatically anonymous under the GDPR, which sets a stricter test. De-identification lowers privacy risk but does not override contracts. Records a customer agreement forbids a company to share remain forbidden after the names are removed. Outrich Data de-identifies supplier data before any buyer sees it and documents the method.
The method should be written down as it is applied: which identifier classes were removed, with which tools, how free text was reviewed and what the sample testing showed. That record goes into the data card.