Skip to content

Manual Review Workflow

Definition

When the decision engine routes a verification to manual review, a human agent examines the case and makes a final approve/reject decision. Efficient manual review is critical for operations at scale.

Review Queue Architecture

graph TD
    A[Cases Needing Review] --> B[Priority Queue]
    B --> C{Assignment}
    C -->|Round-robin| D[Agent Pool]
    C -->|Skill-based| E[Specialized Agents]
    C -->|Priority-based| F[Senior Agents for high-risk]

    D & E & F --> G[Agent Review Interface]
    G --> H[View document images]
    G --> I[View AI scores + flags]
    G --> J[Compare face match]
    G --> K[Check screening results]

    H & I & J & K --> L{Agent Decision}
    L -->|Approve| M[✅ Approved + reason]
    L -->|Reject| N[❌ Rejected + reason]
    L -->|Escalate| O[Senior reviewer]

Agent Interface Requirements

Component What It Shows
Document viewer Original images with zoom, rotate, enhance
AI overlay Highlighted issues — low-confidence OCR fields, forensic flags
Face comparison Side-by-side selfie vs document photo with match score
Screening results PEP/sanctions hits with source details
Customer data Extracted OCR data, editable for corrections
Decision buttons Approve / Reject / Escalate with mandatory reason selection
Timer SLA countdown for the case

SLA Targets

Priority Target Resolution Case Type
P1 - Critical < 5 minutes Sanctions escalation
P2 - High < 30 minutes Borderline matches, PEP hits
P3 - Normal < 4 hours Standard review cases
P4 - Low < 24 hours Bulk re-review, quality audit

Agent Productivity

Metric Target
Cases per hour 10-15 (standard), 5-8 (complex/EDD)
Accuracy > 98% agreement with quality audit
SLA compliance > 95% within SLA

Key Takeaways

Summary

  • Manual review handles 8-15% of verifications — must be efficient and scalable
  • AI-assisted review (pre-highlighted issues, scores) dramatically improves agent speed
  • SLA management with priority queuing prevents backlogs
  • Audit trail for every manual decision is required for compliance
  • At scale (1M+ verifications/month), this means 100K-150K manual reviews — operational challenge