Entity extraction is the operation that converts free-form speech ("Tuesday at three with Dr. Martinez") into typed values (date=2026-05-26, time=15:00, provider=martinez_g). Every meaningful action the agent takes downstream depends on entities being parsed correctly.
The accuracy bar is higher than it looks. A booking with the wrong date is worse than no booking at all, the customer thinks they are scheduled and shows up on the wrong day. The agent has to verify ambiguous entities ("Tuesday could be tomorrow or next week, which did you mean?") rather than guessing.
Vertical-specific entity extractors are the production-grade pattern. A clinic agent extracts insurance carriers and provider names; an auto-shop agent extracts VINs, make and model, and symptom descriptions; a dispatch agent extracts addresses and equipment types. A general-purpose extractor will miss vertical vocabulary that the vertical extractor catches.
Vorel ships vertical-specific extractors, the VIN parser knows the seventeen-character format and check digit; the address parser knows the local postal conventions for the customer's region.