Key takeaways
- Capture quality drives accuracy more than anything else.
- Run liveness and matching on the same selfie.
- Route the grey zone to humans instead of auto-rejecting.
- Keep decisions and scores; delete templates on schedule.
1. Capture the ID
Guide the user to fill the frame, avoid glare and hold still. Reject blurry or cropped captures immediately with face detection rather than failing later.
2. Capture the selfie
Use the front camera with an oval guide, good light and a neutral expression. Show real-time feedback like "move closer" or "too dark".
3. Check liveness
Run liveness on the selfie before trusting it. For higher-risk products, add device-integrity signals to resist injection attacks.
4. Match selfie to ID
Call /v1/verify with both images. Set your threshold using test data from your own customer base, and consider stricter thresholds for high-value accounts.
5. Handle the grey zone
Define three bands: auto-approve, manual review and decline. Manual review catches aged ID photos, glare and genuine edge cases without losing good customers.
6. Accessibility and fairness
Offer an alternative path for users who can't or won't use a camera. Measure approval rates across demographic groups where lawful, and investigate gaps.
7. Records and retention
Store the decision, score, timestamp and consent record. Delete biometric templates when the purpose ends, in line with your retention schedule and local law.
