What is an AI receptionist, and what does it do all day?
An AI receptionist is a computer program that answers your phone and talks with the caller. It is not a menu and not a recording. It listens, works out what the caller wants, and does something about it: answers a question, books a time, takes a message, or passes the call to a person. The easiest way to understand one is to follow it through a working day, from the first ring to the last summary.
The first ring: a greeting and a question
Every call starts the same way. The agent picks up, says the name of your business, and says it is an AI. That last part matters. Callers relax when they know what they are talking to, and they get cross when they find out late. Then it asks how it can help, and it waits.
Most callers say what they want in a sentence or two. Some take longer. The agent's first job is to find the request inside the words. "I think I need to move my Thursday thing" is a rescheduling. "Are you open on Saturday?" is a question it can answer from your opening hours.
Under the hood, three things happen in a loop. Listening turns speech into text. Thinking is a language model reading those words together with your instructions. Speaking turns the reply back into a voice. Some engines fold all three into one model, which sounds more natural and costs more per minute. We compared the two approaches in Pipeline vs. speech-to-speech.
The middle of the day: the four jobs it does most
Listen to a week of calls at almost any business and the same requests come back again and again. A receptionist, human or not, spends most of the day on four of them.
- Answering questions from your own material: hours, prices, directions, what to bring. The agent quotes from the documents you gave it and says so when it does not know.
- Booking, moving and cancelling appointments through a calendar tool. It only says "booked" once the calendar has confirmed.
- Taking messages and callback requests: a name, a number, what it is about, and when a good time to call back would be.
- Handing over to a colleague with a warm transfer, a cold transfer, or a written message.
The booking job is the one people ask about most, so it has its own post: How AI appointment booking by phone actually works.
The hard calls: knowing when to stop being clever
Some calls are not for a machine. An upset customer. A medical question. A caller who is grieving, or scared, or shouting. A good agent has one rule for these: get a person, or take a message and promise a callback. It does not argue, guess, or improvise a policy you never wrote.
This is where products differ most. A basic setup lets the language model say whatever sounds plausible, which is how an agent ends up promising a refund nobody approved. A more careful setup adds a second model that reads every reply before it is spoken and blocks claims the agent cannot back up. In NordTell that is the Hybrid engine, described under engines. Whatever product you use, ask how it stops the agent from bluffing.
After the call: the paperwork
When the caller hangs up, the agent's day is not over. It writes a transcript, a short summary and an outcome, such as "booked", "message taken" or "transferred". It can tell your other systems what happened through a webhook, so a new lead lands in your CRM without anyone typing. The recording is on, off, or on only after the caller agrees, whichever you chose for that agent. Calls, recordings and transcripts expire on the schedule you set.
The result is a list of calls you can read over coffee the next morning: which ones were booked, which ones need a call back, and which ones the agent was not sure about.
Do you need one? A three-question test
First, be clear about what an AI receptionist is not. It is not a phone menu; the caller talks, and it adapts. It is not a replacement for judgement; the tricky calls still go to people. And it is not magic; it knows only what you tell it, so the instructions you write are its job description. Then ask yourself three questions.
- Do calls go unanswered? Count them for two weeks before you guess. Here is a method.
- Are most calls the same handful of requests?
- Is there a person the agent can hand the odd calls to?
Three yeses mean it is worth a trial. Build it in the browser first and talk to it with a microphone. Connect a phone number only when you would be happy for a real customer to hear it.
Where NordTell fits
NordTell is a place to build this kind of agent. You write its instructions in plain words, pick a voice and a language, test it in the browser, and connect a number you buy or bring. It runs on servers in the EU and is built in Copenhagen. Every package includes every feature; you pay for minutes, not seats.