What an AI virtual PBX is
A virtual PBX is a phone system that does not sit in a cupboard in the office but on a server you reach over the internet: you configure it from the browser, phones connect from any site and there is no hardware to buy or repair. If you want the full comparison with an on-site system, you will find it in the guide to the cloud PBX and how to choose one.
The "AI" part is a recent addition. Alongside the classic components (keypad menus, queues, opening hours) there is an AI voice agent: a synthetic voice driven by a language model that listens to the caller, understands a request expressed in ordinary words and decides what to do, instead of forcing them to press keys.
The PBX is not the phone line. To receive and make calls it needs numbers and lines supplied by an operator, connected through a SIP trunk or a SIP account. This distinction comes up again and again in this guide: the PBX decides where the call goes, the operator carries it into and out of the telephone network.
The building blocks: what each one does
A modern virtual PBX is built by combining a few blocks. Each has a precise job, and the quality of the service depends above all on the order you put them in.
The AI voice agent
This is the block that holds the conversation. You tell it in plain language who it is, what it knows about the business and what it has to achieve: answer frequently asked questions, collect the caller's name and reason for calling, book an appointment, pass the call to a colleague. The most delicate part is writing good instructions, which is why there is a dedicated guide on how to write instructions for an AI voice agent.
In the AI voice agent of our partner robottino.ai's virtual PBX, for example, the actions the assistant can take are defined one by one: booking on a calendar, sending the collected details to a business application, routing the call to a queue or an agent.
The IVR menu
IVR (Interactive Voice Response) is the classic keypad menu: "for accounts press 1, for support press 2". It is less flexible than an AI voice agent, but it is predictable, quick for frequent callers and it never misunderstands a key. Many businesses use the two together: the IVR to sort calls, the AI for callers who do not know which option to choose.
Call queues
A queue keeps callers waiting in an orderly line when all agents are busy, with a message and music, and passes them to the first one who becomes free. You decide how calls are distributed (all at once, in turn, by priority) and what happens if the wait gets too long: a message, a callback, another number.
Time conditions
These are your opening hours, written down once, with breaks, closing days and public holidays. The PBX checks them on every call: during opening hours the call goes to the agents, outside them it goes to a message, to voicemail or to the AI voice agent.
Transfers
A transfer takes the call out of the PBX, to a mobile or to a number at another site. It is the simplest way to involve people who do not have an extension, such as the owner on a business trip or the on-call technician. Remember that the leg to the external number is an outbound call and is charged as one.
The agent panel
This is the workstation of the person answering: a phone in the browser with information about the caller alongside it. The advantage with AI is that the agent sees what the customer has already told the voice agent, and does not have to make them repeat everything from the start.
Call recording and analysis
Recording saves the audio of conversations for quality control or as evidence of an agreement. AI analysis goes one step further: it transcribes conversations and classifies them according to criteria you write, for example "find customers who asked for a quote and did not receive one". Before switching recording on, read the guide on the rules for recording calls.
The journey of a call, step by step
Take an engineering practice with six people: two at reception, three engineers and the owner. The public number is 06 1234 5678. This is what happens when a client calls on a Tuesday at 11 am.
- The client dials the number. The telephone network delivers the call to the operator that manages the number, which passes it to the PBX over the SIP line.
- The PBX checks the time condition: on Tuesday at 11 the practice is open, so the call continues along the "open" route.
- The AI voice agent answers: "Rossi Engineering, good morning. How can I help you?" The client explains that they need to move a site visit.
- The voice agent understands that the request concerns appointments, asks for the name and the date of the visit and confirms the details by repeating them.
- Because the change has to be agreed with an engineer, the voice agent passes the call to the reception queue.
- Both people at reception are busy: the client waits in the queue with a message. After 40 seconds one of them is free and her phone rings.
- On the panel the receptionist sees the summary ("Bianchi, move Thursday's site visit") and answers already knowing what it is about.
- If the call is recorded, the audio goes into the report at the end of the conversation, where it can be played back or analysed.
At 8 pm the same number follows a different route: the time condition says "closed", and the AI voice agent takes the request and emails it to the practice or, if it is an emergency covered by the instructions, transfers the call to the mobile of whoever is on call.
What you need to get started
The list is shorter than it seems. Almost everything is configured from the browser.
- A phone number your customers call: a new virtual number or the number you already have
- A SIP line that brings that number's calls to the PBX and carries your outbound calls
- The virtual PBX, with the AI voice agent, menus, queues and hours configured
- The workstations: IP phones, softphones on computers or mobiles, or the panel in the browser
- A stable connection: a G.711 call uses about 100 kbit/s in each direction
- The voice agent's instructions, written and tested before you put it in front of customers
For the first point you can choose from virtual numbers in 232 Italian cities and more than 90 countries, with inbound calls included in the monthly fee. For the second, the PBX connects to the operator through a SIP trunk or a registered SIP account: this is the part Vohippo takes care of, while the PBX is a separate product.
How long it takes
For a four-person office, a basic setup (hours, voice agent, one queue, voicemail outside hours) takes a few hours of work, plus the time needed to test the AI's instructions. If the number is new, allow for the document check, which usually completes in 1-3 working days.
The limits to know beforehand
An AI voice agent is not infallible. It can mishear a name, mix up two dates or answer confidently a question it does not know the answer to, unless the instructions forbid it. That is why every flow must have a way out to a person or to voicemail.
Background noise, phones on speaker in the car and very strong accents make understanding worse. Line quality matters too: a call with choppy audio is hard for a person and even harder for an automated system.
Then there is privacy. The voice agent collects personal data and the conversation is often transcribed or recorded: callers must be clearly informed, with a message at the start of the call, and in the EU the data must be processed in line with Regulation (EU) 2016/679 (GDPR). For the details of your specific case, ask your privacy adviser.
A full PBX or a simpler AI secretary?
Not everyone needs queues, menus and panels. If you work alone or with one other person and simply want no call to go unanswered, the Vohippo AI secretary may be enough: a step in the number's routing that answers, collects name, contact details, reason and urgency, and sends you a card with the summary and the «Accept» and «Reject» buttons. If you accept, the call continues to your phone.
When instead you need calls shared among several people, opening hours, managed waiting and per-agent statistics, you need a real PBX. One option is the virtual PBX of our partner robottino.ai, which brings together AI voice agent, menus, queues and time conditions; Vohippo remains the operator supplying the numbers and lines.
Frequently asked questions
What is the difference between a virtual PBX and a cloud PBX?
In practice there is none: they are two names for the same service, a PBX hosted on remote servers and managed from the browser. "Virtual" stresses that there is no physical box, "cloud" that it lives on the internet.
Does the AI voice agent replace people?
No, it takes the repetitive calls off their hands and prepares the important ones. Complex requests, negotiations and unhappy customers must reach a person, and a good flow always includes that way out.
Can I keep my current number?
It depends on how the number is connected. The PBX needs to receive calls on a SIP line: with Vohippo you can buy a new virtual number or, if your number is already on a VoIP line with another operator, register it as an «external number» without changing operator.
Does the AI voice agent understand dialects and foreign languages?
Current systems understand speech with regional accents and many foreign languages well, but strong dialect is still difficult. It is worth stating in the instructions that if it does not understand after two attempts, it should pass the call to a person.
Do I need a technician to set it up?
Not for a basic setup: hours, menus and the voice agent are configured from the PBX's portal. You need a technician if you want to connect the PBX to a business application or if you use desk phones that have to be configured one by one.