AI phone assistant

Technically, three building blocks work together: speech recognition turns what was said into text, a language model words the reply, and speech synthesis speaks it. Newer systems process speech directly, without the detour through text. On top of that come the telephone connection, a knowledge base with the information of the business, and interfaces to a calendar or CRM.

Typical tasks are information on opening hours and services, appointment booking, taking down requests for a callback and transferring to a human. Unlike a keypad menu, the assistant does not follow a fixed tree but responds to freely worded sentences. It reaches its limits where knowledge is missing or where a decision is reserved for a person.