Calls that repeat and have a clear finish: booking and rescheduling, order status, first-line support, qualifying a lead. Calls with high stakes or strong feelings are better with a person from the start, and we'll say which of yours are which.
How is it different from a phone menu?
A phone menu makes callers press numbers through fixed options. A voice agent understands what the caller says, asks a follow-up when it needs one, and does the task — books, looks up, updates — or passes the call to the right person.
Will it work with our phone system, calendar and CRM?
Usually, yes. It connects to your phone numbers or SIP line, and to your calendar, CRM, help desk and order system through their APIs, each with only the permissions it needs. In an app or on a website it runs behind a talk button instead of a number.
Will callers know they're talking to an AI?
They should, and we set it up that way: the agent says so at the start, wherever the law requires it and wherever you prefer. A natural voice and quick replies keep the call easy, and a person is always one request away.
What happens when it can't help?
It says so and transfers the call to your team with a summary of what was said and done, so the caller doesn't repeat themselves. Out of hours, it takes a message or books a callback.
Which languages can it speak?
Most of the languages your callers use, English, Spanish and Portuguese among them, switching when the caller does. Voices and pronunciation are tested in each language before launch.
How fast does it reply?
Fast enough for a phone call: usually under a second from the end of the caller's sentence. Speech is recognised while it's spoken, and the agent stops talking when the caller interrupts.
Does it follow recording and consent rules?
We set recording notices and consent for each country, keep card numbers and other sensitive details out of transcripts, and keep audio and transcripts in the region you choose. Outbound calls follow the calling rules where they're made, such as PECR in the UK.
Do you build on Vapi or Retell, or from scratch?
Either. Vapi and Retell get a standard call flow live quickly; LiveKit or Pipecat on your own infrastructure give full control over the models, the speech, the speed and the data. We recommend one after looking at your calls, and the agent stays yours either way.
How long does it take, and what does it cost?
Choosing the calls and designing the conversation takes one to two weeks, a working agent on simulated calls two to six, and a pilot on real calls a few weeks more. Running cost is per minute of calls — speech, voice and model — which we estimate before we build. After a first call we send a written estimate.