What is Equal Error Rate (EER)?

Equal Error Rate (EER) is the point on a biometric or classifier curve where false accepts equal false rejects. One number, quick to compare across engines and datasets; lower is better. Teams use EER to benchmark models, capture setups, and versions without arguing about an operating threshold—useful while exploring.

Still, business rarely runs “at EER.” Border control accepts fewer false accepts; consumer apps hate false rejects. So you pick a threshold that matches risk appetite, then watch drift over time. Segment results by environment, device class, and demographics; a “great” global EER can hide weak pockets and fairness issues.

Operationalize it properly: track DET/ROC curves, lock evaluation protocols, store artifacts so audits aren’t archaeology. When deepfakes get better or fraud pressure rises, push thresholds tougher and add stronger evidence—robust liveness checks and end‑to‑end identity verification. EER is a lighthouse, not the harbor—steer by it, then dock where your product and liability demand.