Cost Optimization¶
Definition¶
Optimizing the cost per eKYC verification — GPU compute, storage, API calls, and operational overhead.
Cost Breakdown (Typical)¶
| Component | % of Cost | Optimization |
|---|---|---|
| GPU compute | 40-60% | INT8 quantization, spot instances, right-sizing |
| Third-party APIs | 15-25% | Screening API, database verification |
| Storage | 10-15% | Lifecycle policies, embedding vs raw images |
| Manual review | 5-15% | Improve STP rate, AI-assisted review |
| Infrastructure | 5-10% | Reserved instances, auto-scaling |
Cost per Verification Benchmarks¶
| Scale | Cost/Verification | Notes |
|---|---|---|
| 1K/day | $2-5 | Low volume, high fixed costs |
| 10K/day | $0.50-2 | Moderate economies of scale |
| 100K/day | $0.10-0.50 | Good GPU utilization |
| 1M/day | $0.05-0.20 | Maximum economies of scale |
Key Takeaways¶
Summary
- GPU compute is the largest cost — INT8 quantization + spot instances can reduce by 60-80%
- Increasing STP rate reduces manual review cost — the most expensive per-unit component
- Storage lifecycle (hot → cold → delete) prevents cost from growing linearly with history
- At scale (1M+/day), cost per verification drops to $0.05-0.20 — massive economies of scale