PII and Anonymisation Checks Before Onboarding

A dataset advertised as "anonymised credit card transactions" or "de-identified geolocation pings" can still let a sophisticated user re-identify individuals if enough other fields — merchant, timestamp, rough location — are left intact. Before onboarding any new alternative dataset, quant research teams run a PII and anonymisation check to confirm the data actually is what the vendor claims, both for legal exposure and because re-identifiable data can carry regulatory and reputational risk that outweighs any alpha it offers.

What the check covers

The review looks for direct identifiers (names, emails, account numbers, device IDs) that should never appear, and for quasi-identifiers — combinations of fields like zip code, birth date, and gender, or precise timestamp plus location — that can re-identify a person even without any single field being obviously personal. It also checks whether the vendor's anonymisation method (hashing, aggregation, differential privacy) is applied correctly, since a poorly salted hash of an email address can often be reversed by brute-force matching against known email lists.

Worked example

A geolocation vendor claims its foot-traffic dataset is anonymised because device IDs are hashed. The review finds that each hashed ID still has a consistent home-location cluster (the device's overnight pings), which combined with public property records could re-identify a large share of individuals — the hash conceals the ID string but not the underlying movement pattern, so the dataset fails the check and is either rejected or requires additional aggregation before use.