Face Duplicity check (1:N)
Face duplicity check searches face from the new digital identity record against all the existing Customer records — a 1:N search. It answers a fraud-and-identity question that document checks alone cannot: have we seen this face before, and under what identity?
What it detects
- Duplicate accounts — the same person onboarding again with a different document.
- Name change - the same person onboarding again with a changed name
- Fraudsters — a fraudster with multiple forged ID documents.
How it works
- When a workflow reaches the
dedup_facestep, the face from the digital identity is searched against the watchlists — the Customer watchlist, the Blocklist, the In-progress watchlist of verifications running right now, and the Review watchlist of identities waiting for an operator. - The search returns up to 10 candidates, re-ranked by the similarity score and the watchlist it came from.
- Every candidate is also compared 1:1 by document data (name, date of birth, numbers) using the rules of the document duplicity check, so each hit carries a face score and a text score.
- The top result and its score determine the step result; the combined duplicity outcome then merges it with the document search.
Outcomes
dedup_face.result distinguishes where the match was found:
| Result | Meaning |
|---|---|
no_match | The face is new. |
customer | Matches an existing Customer — the attempt can be merged or sent to review. |
blocklist | Matches a blocked person — the attempt is rejected. |
concurrent | Matches another verification happening right now (see below). |
review | Matches an identity that is waiting for manual review. |
Concurrent-onboarding protection
Two sessions for the same person running at the same time could each create a separate Customer. To prevent this, the face of an in-flight applicant is briefly tracked while its verification is undecided, so a simultaneous second attempt is flagged as concurrent rather than slipping through as unique. Once the verifications resolve, this temporary tracking is cleaned up automatically.
Combining with document data
Face duplicity check is most powerful alongside document duplicity check: comparing the face signal with the name + date of birth signal is what separates a returning customer from a stolen ID, a twin, or a namesake. The best_duplicate step produces a single combined outcome from both; see also the matrix in Identity Verification.