Direct identifiers, such as a name or an account number, point to one person. Quasi-identifiers work together. Many people share a birth year, and many share a town, but few share a birth date, a town, a profession and an employer. When those facts are also available elsewhere, for instance in a public profile, a record can be linked to a name without any direct identifier being present.
Deciding which fields are quasi-identifiers is a judgment about what an outsider could plausibly know. Dates of events, locations, roles, team names and unusual attributes are typical candidates. In workplace data, a combination of department, seniority and start date can be enough. Once identified, quasi-identifiers are treated by generalizing them, banding them, or dropping those that add little value. Their effect is measured with k-anonymity, which counts how many records share each combination. Free text can contain quasi-identifiers too, in phrases that describe a person's role or circumstances, and these need review by sampling.
The list of quasi-identifiers should be recorded with the de-identification method. A buyer reviewing the data card can then see which fields were considered and how each was treated.
Example
A support record lists no name, but shows the caller is the head of finance at a named regional hospital. Role and employer together identify one person.