Financial AI: Identity and Risk

Connect identity checks, business verification and financial risk review in one workflow. Collect evidence at onboarding, send exceptions to a reviewer and retain the decision for audit.

Illustration of an analyst reviewing an identity match and a flagged transfer

KweGo Project Scope

The agreement specifies CITIA’s work on identity verification (KYC), business and beneficial ownership checks (KYB), transaction risk analysis (AML) and supporting model controls.

  1. Capture documents
  2. Check evidence
  3. Review exceptions
  4. Record decisions

What reviewers see

A case record is designed to show what triggered the flag, such as a mismatch between the name printed on an identity document and the name in its machine-readable zone.

Review and Access Controls

The implementation plan gives authorized reviewers access to the evidence behind each alert and requires human override. Any deployment is subject to applicable regulatory requirements and acceptance testing.

Illustration of reviewers comparing identity records, transaction evidence and a checklist

Multiple checks in one workflow

Combine document checks, face comparison and liveness checks where required. Route incomplete or inconsistent evidence to manual review.

Evidence linked to each decision

Keep the check results, source records and reviewer decision together so another authorized reviewer can retrace the case.

Verification, Risk and Model Governance

Each function must pass its agreed acceptance tests before it can be used in production.

Illustration of a person photographing an identity card with a phone

Identity Verification

Use OCR and machine-readable zone (MRZ) checks to extract and compare document data. Biometric comparison and liveness checks can flag cases for manual review, with the evidence and decision recorded in the case history.

Document capture
Document checks start with image readability. Where a machine-readable zone is available, its fields are compared with the printed fields. Missing fields or mismatches go to a reviewer.
Face match and liveness
Where the workflow requires it, compare the applicant’s capture with the document portrait and use liveness checks. Route inconclusive results to a reviewer.
Illustration of a reviewer comparing a corporate registration record and an ownership chart

Business Verification (KYB)

Compare entity records with registry information, review beneficial ownership, and route sanctions matches and enhanced due diligence findings into case management.

Entity and ownership checks
Compare legal name, registration information and ownership declarations with available records. Record gaps in beneficial ownership evidence for follow-up.
Due diligence review
Link potential sanctions matches and enhanced due diligence findings to the case. Keep the reviewed evidence and disposition together.
Illustration of an analyst tracing flagged transfers on a transaction network

AML Risk Intelligence

Monitor transaction patterns, use graph analysis to trace relationships, and prioritize anomalies for analysts to investigate and record their decisions.

Transaction patterns
Group related activity and surface unusual sequences for review. Show the records, dates and relationships used to raise the alert.
Investigation and escalation
Prioritize alerts and give analysts a place to record findings, supporting evidence and escalation decisions.
Illustration of reviewers checking model validation results and version history

AI Governance

Evaluate models against agreed test cases, test for drift, capture reasons for model outputs, and track versions, approvals and overrides in an audit trail.

Evaluation before release
Check performance on normal cases and exceptions. Track the model version, test cases and reviewer sign-off.
Human review and overrides
Keep the evidence behind alerts and model outputs accessible to authorized reviewers. Record corrections, overrides and the final case decision.

Open Finance Data Access

These controls describe how a separate open finance integration can handle customer permission and data sharing.

Illustration of a customer choosing which bank accounts to share with an app

With the customer’s permission, an integration can pass the authorized account fields to an application or an analyst for a specified purpose.

Consent and freshness
Check the customer’s grant for each application before delivering requested fields. Reuse eligible data only within the grant and applicable freshness requirements.
Revocation and audit
Stop new requests and cached deliveries when consent is withdrawn. Record the application, authorized fields and time of access or revocation.

Use Cases

Tiered identity checks

Define which checks apply to routine onboarding, missing evidence and higher-risk cases. Add face comparison or manual review where the case calls for it.

Business onboarding and due diligence

Compare submitted information with registry records, trace beneficial owners and send unresolved discrepancies to a reviewer.

Transaction alert review

Show the reviewer permitted transactions, timestamps and the reason for the alert. Test that a balance-only application cannot retrieve transaction details and that revocation stops further delivery.

  1. Onboarding checks
  2. Evidence review
  3. Case disposition