Workflow Automation · Read + Write

KYC document verification

Automate Rep · High Reason · Medium

The problem

Compliance teams manually verify ID + address + selfie for every new account. Days of backlog; high attrition for users abandoned mid-onboarding.

The AI approach

OCR the ID and address proof; match selfie against ID photo; check expiry + sanctions list; apply KYC tier rules; auto-approve clean cases.

The outcome

~90% touchless

Onboarding from days → 2 minutes for clean cases. Compliance team focuses on edge cases and watchlist hits.

Try itInput → Process → Output

Input — ID + address + selfie
Passport · ID
👤
RITBUMROONG, THANACHART
P · TH · DOB 1980-04-12
passport no. AA1234567 · exp 2031-08
Selfie · liveness
😊
live · captured 2s ago
Address proof
Utility bill · TRUE Mobile
123 Sukhumvit Rd, Bangkok 10110
Issue date: 2026-04-15
Process — AI pipeline
1OCR ID + address proofReadPredictive
2Face-match selfie ↔ ID photoReadPredictive
3Apply KYC tier + sanctions rulesRulesSymbolic
4Decision: approve / review / rejectWriteSymbolic
Output — verification result
Click Run demo to verify the documents and apply KYC rules.
KYC verification PASS · Tier 2
✓ID document type — Passportaccepted (TH)
✓ID expiry2031-08 · valid
✓Face match (selfie ↔ ID photo)96.4% · pass
✓Livenesspass · 0 anomalies
✓Sanctions / PEP matchno hits
✓Name on address proofmatch
✓Address proof recency (≤ 90 days)19 days
Decision: Approve · Tier 2 (full account). All 7 checks passed; no manual review required.

Three AI types in this use case

SymbolicKYC tier rules (age, country, doc type required); sanctions / PEP / watchlist matching; ID expiry + name-on-address-proof check.
PredictiveOCR + ID layout classifier; face-match similarity score; document-tampering detector.
GenerativeLLM normalises name variants and address formatting across docs (Latin / non-Latin scripts).

The stack

  • IDP · Onfido / Jumio / Persona
  • Face-match · liveness + similarity
  • Sanctions / PEP · WorldCheck / Refinitiv
  • Rules engine · per-jurisdiction KYC

When this works

  • Doc types are limited (passports, national IDs in supported countries)
  • Sanctions / PEP feed is current
  • Manual review queue exists for the 10% edge cases

When it fails

  • Spoofed liveness with deepfake selfie
  • Address proof in unsupported language → OCR fails
  • Watchlist false-positive for common names
  • Auto-reject without review path → compliance discrimination