Obtain from your IDV vendor the documentation describing their face match score range, the recommended default threshold, and the FAR and FRR curves for their model
Set an initial threshold value in the vendor SDK or API configuration, erring on the side of the vendor's recommended value before tuning
Instrument your pipeline to log every face match score, the pass or fail outcome, and downstream fraud or dispute indicators for each transaction
After accumulating sufficient volume, analyze the score distribution to identify a threshold that balances your acceptable false accept rate (fraud risk) against your false reject rate (user friction)
Implement threshold changes as a configurable parameter updated without a code deploy so that you can respond quickly to observed fraud patterns
Periodically audit the match outcomes by demographic group to identify and address any disparate impact in false reject rates across age, gender, or skin tone
Known gotchas
Face match thresholds interact with the liveness check; lowering the face match threshold to reduce friction without a corresponding liveness improvement increases spoofing risk
Vendor model updates can silently shift the effective FAR and FRR curves even for the same nominal threshold value; monitor your score distributions after any vendor model version change
A single global threshold may not be appropriate for all document types or acquisition conditions; consider separate thresholds for passport photos versus driving license photos if the vendor supports it
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp