In this guide8 sections
- 01What does an AI voice agent actually do on the phone?
- 02How is it different from an IVR, a menu robot or a chatbot?
- 03What is an AI voice agent made of?
- 04A few glossary terms in plain English
- 05What can it do well today, and what not?
- 06Where does an AI voice agent earn its place in a business?
- 07How do you choose an AI voice agent? Honest criteria
- 08In short: where do you start?
An AI voice agent is software that holds a phone conversation in place of someone on your team. It calls customers or answers when they call, understands what they say in their own words, replies out loud and then does what a colleague would do behind the scenes: takes a note, books a slot, sends a message or updates a record. It is not a “press 1” menu, and it is not a chat window with a voice bolted on. It is a real conversation with a beginning, a middle and an end, and it leaves a clear outcome behind.
This guide explains what an AI voice agent is made of, how it differs from tools you already know, what it can and cannot do today, and how to choose one with your eyes open.
What does an AI voice agent actually do on the phone?
It works in two directions:
- Initiated calls (outbound): the agent calls a list of people with one clear purpose. It confirms an order, reminds someone of an appointment, asks for a missing document or runs a short survey.
- Received calls (inbound): the agent answers your business number, including at night and at weekends, says who it is, finds out what the caller wants and either resolves it or notes the request for a colleague.
Every call leaves something useful behind: a recording (where enabled, with the caller informed), a transcript, a summary and an outcome such as “order confirmed”, “address corrected”, “booked for Thursday” or “caller asked for a colleague to ring back”. That outcome can then be written to your CRM or ERP.
Beyond the voice itself, the agent can send an SMS, a WhatsApp message or an email after the call with whatever was agreed. You can follow the full flow on the how it works page.
How is it different from an IVR, a menu robot or a chatbot?
These tools get mixed up all the time, but they solve different problems.
| Criterion | IVR or menu robot | Text chatbot | AI voice agent |
|---|---|---|---|
| How the person communicates | Presses keys or says words from a list | Types | Speaks freely, in normal sentences |
| What it understands | Predefined options | Text, sometimes with predefined intents | The meaning of the sentence, even when the person corrects themselves or changes topic |
| Conversation path | Fixed, like a tree | Depends on setup | Flexible, within your rules |
| Can it call first | Usually not | No (it is not voice) | Yes |
| What it leaves behind | An option number | A chat log | Recording, transcript, summary, outcome |
| Best for | Simple routing to departments | Written questions on a website | Phone conversations with a clear purpose and a clear end |
The underlying difference is simple. With a menu, the person adapts to the system. With an AI voice agent, the system adapts to the person. If someone answers “well, I’m not home, but my neighbour will be”, a menu gets stuck. A well-configured agent understands that the address still stands and the parcel can be handed to someone else, then asks what it needs to record it correctly.
If you mainly want to compare it with classic answering machines and phone menus, read IVR, automated phone system or AI voice agent.
A chatbot is still useful for anything that can be solved in writing. The phone remains the channel many people prefer, especially for urgent matters, for confirmations, or when they simply do not feel like typing.
What is an AI voice agent made of?
Think of it as a new employee with four parts working together.
1. Ears and mouth: the voice layer
Two technologies sit here, and you will meet them in every glossary.
- Speech to text (STT) turns what the person says into text. Its quality matters enormously: if “12 Poplar Street” is heard as “2 Poplar Street”, the rest of the conversation starts from the wrong place.
- Text to speech (TTS) turns the agent’s reply into a voice. This decides whether it sounds natural or like an old switchboard recording.
2. The brain: understanding
This is the part that chooses what to say. It receives the text, remembers what was said earlier in the call and decides the next step. A good agent does not make things up. It answers from the documents and data you provided and, when it lacks the information, says so.
3. The rules: what it may and may not do
This is your control panel. You define the purpose of the call, how it opens, which questions it asks, which information it may share, and when it stops and hands over to a person. The calling rules live here too: calling hours, how many retries, and who must never be called again.
4. The hands: integrations
An agent that only talks produces transcripts. An agent with integrations produces results. It reads the order or the case file before the call, writes the note afterwards, finds a free slot in the calendar and sends the after-call message. See what can be connected on the integrations page.
A few glossary terms in plain English
- Latency: the pause between the moment the person stops talking and the moment the agent starts replying. Long pauses make a call feel awkward. Our agent replies almost instantly in conversation, in under one second.
- Transcript: the full text of the call, line by line. Reading it is faster than listening to the recording.
- Call cadence: the rule for retrying when nobody answers, meaning at what times and how many times.
- Knowledge base: the documents, texts and web pages the agent takes its answers from. How to prepare one is covered in the guide to the AI voice agent knowledge base.
- Handover to a person: the moment the agent deliberately stops because the topic goes beyond its rules, and passes the case on.
- Inbound and outbound: received calls and initiated calls.
What can it do well today, and what not?
Honestly, an AI voice agent does not replace your whole team. It is very good at repetitive conversations with a clear goal and a limited number of variations. It struggles with anything that needs deep empathy, negotiation or judgement in an unexpected situation.
It works well for:
- confirmations and reminders (orders, appointments, deliveries);
- answering calls outside business hours and noting the request;
- bookings, offering two or three free slots;
- requests for missing documents or information, followed by a written message with the exact list;
- short surveys after a delivery or a visit;
- frequent questions, when the answer exists in your company documents.
It stays with people:
- emotionally charged complaints or delicate situations;
- negotiations, exceptions and decisions with important consequences for the customer;
- any topic for which you have not given the agent information or rules;
- cases where the person explicitly asks for a human, in which case the agent notes the request for a colleague.
One more thing worth knowing: an agent can make mistakes, like any new colleague, especially if the information it receives is incomplete or contradictory. That is why the test before launch matters, along with listening to calls in the first days and refining the instructions. For how to judge quality, see how to measure an AI voice agent.
Where does an AI voice agent earn its place in a business?
A few typical scenarios, without claiming they are the only ones:
- Online shop with cash on delivery: calls every new order and confirms the address, products and amount before the parcel ships. See order confirmation for cash on delivery.
- Clinic or practice: answers the phone when reception is busy and books appointments into the calendar.
- Business that works with case files: calls clients whose documents are missing, then sends them the list by WhatsApp and email.
- Any business that gets calls after hours: takes the request, records it and leaves it for the morning.
More examples are collected on the scenarios page. A simple test: if your people repeat the same conversation dozens of times a day, almost word for word, you have found a good candidate.
How do you choose an AI voice agent? Honest criteria
Before you look at prices, look at what you can check yourself.
- Can you hear it on your own case? A demo on a generic scenario says little. Ask for a test on your own phone, with your real lines.
- Does it speak your customers’ language well? Not just “supports the language”, but sounds natural and copes with accents and proper names. Check the available languages.
- Does it answer only from what you gave it? Ask what it does when it does not know. A good answer is “it says it does not know and notes the request”.
- What does it leave behind after the call? A recording with the caller informed, a transcript, a summary, a structured outcome.
- Does it connect to your systems? Get clear confirmation for your CRM, ERP, calendar and messaging channels.
- How is data handled? Where it is hosted, who is the controller and who is the processor, how “do not call me again” is respected. See the security page.
- How much do you control? Calling hours, retries, scripts and the moment of handover to a person.
- Do you know what you will pay? Ask for the billing model, the rounding rule and the setup costs. The details are in how much an AI voice agent costs.
If a provider will not let you test, or cannot explain what the agent does when something falls outside the script, you already have your answer.
In short: where do you start?
An AI voice agent is useful wherever a phone conversation repeats itself and has a clear outcome. Start with a single scenario, small and measurable, test it on your own phone, listen to the first calls and only then expand. You can hear an example on the demo page, and if you want to see what the cost looks like for your volume, the sign-up form shows an estimate before you submit anything.