eKYC Monitoring & Observability¶
Definition¶
Monitoring an eKYC system in production — tracking accuracy, latency, throughput, error rates, and compliance metrics to ensure the system performs reliably at scale.
Key Metrics Dashboard¶
| Category | Metric | Alert Threshold |
|---|---|---|
| Volume | Verifications per hour/day | ±30% from baseline |
| Latency | P50, P95, P99 processing time | P95 > 10s |
| Accuracy | STP rate, manual review rate, rejection rate | STP < 70% or reject > 15% |
| Errors | API error rate, SDK crash rate | > 1% error rate |
| Liveness | Spoof detection rate, false rejection rate | Spoof rate > 5% |
| Face match | Average match score, distribution shift | Mean shift > 0.05 |
| OCR | Field-level accuracy, confidence scores | Avg confidence < 0.90 |
| Screening | Sanctions hit rate, PEP hit rate | Unusual spike |
| Infra | GPU utilization, queue depth, memory | GPU > 90%, queue > 1000 |
Model Drift Detection¶
graph TD
A[Production Traffic] --> B[Score Distribution Monitor]
B --> C{Distribution shift?}
C -->|No shift| D[Normal operations]
C -->|Shift detected| E[Alert: Model drift]
E --> F[Investigate: new attack? Data change? Camera change?]
F --> G[Retrain or adjust thresholds]Key Takeaways¶
Summary
- Monitor accuracy metrics (STP rate, rejection rate) as closely as infrastructure metrics (latency, uptime)
- Score distribution monitoring detects model drift before it impacts customers
- Compliance dashboards (screening hit rates, manual review SLAs) are required for audit
- Set alerts on leading indicators — catch problems before they cascade