AI phone secretary: how it works and how to set it up

What the AI secretary is, and what it is not

The «AI secretary» is one of the nine routing modules for inbound numbers. When the call reaches this step, an artificial voice answers with the opening phrase you wrote, follows your instructions and collects four pieces of information: name, contact details, reason and urgency. Then it asks you whether you want to speak to that person.

It is designed for a specific problem: calls that arrive when you cannot answer, which you do not want to lose but do not want to take every time either. A lawyer in court, a technician on a roof, a doctor between appointments: the secretary acts as a filter, and you decide in a few seconds.

It is not, however, an AI-powered PBX. It offers no menu, does not queue calls, knows nothing about opening hours and does not answer customers' questions on your behalf: its job is to collect the details and hand the decision to you. If you need a voice agent that handles conversations on its own, read the dedicated section further down and the guide on the AI virtual PBX.

What happens from the caller's point of view

Put yourself in the shoes of a customer calling a surveyor's office. This is what they hear:

1. A voice answers with the opening phrase, for example: «Moretti Surveyors, this is the automated assistant. The surveyor is busy at the moment: may I have your name and the reason for your call?»
2. The customer replies by voice, as they would to a person. The secretary asks the questions set out in the instructions, one at a time.
3. Once it has the information, it asks them to hold. The customer hears hold music.
4. If the surveyor accepts, the call moves on to the next step, usually the office phones, and the customer keeps hearing the music until someone answers.
5. If the surveyor refuses, or does not decide within the time allowed, the secretary reads the closing phrase you set and ends the call.

The secretary speaks the caller's language: if the customer replies in English or German, it carries on in that language. For businesses that receive calls from abroad this is a real advantage.

What happens from your point of view

While the customer is holding, you receive a notice: an SMS, which is charged, or a notification in the Vohippo app, available on Android only. The notice takes you to a card with:

• a summary in Italian, with name, contact details, reason and urgency;
• the transcribed conversation, if you want to read the exact words;
• the «Accept» and «Refuse» buttons.

The summary currently arrives in Italian, even when the conversation took place in another language. If you tap «Accept», the flow moves on to the next step: typically a «Ring SIP accounts» that rings your phone or the app. If you tap «Refuse», the call is closed politely, and you still have the details to call back whenever you like.

You have a limited time to decide: the «Maximum caller wait», from 30 to 600 seconds, with a default of 120. Two minutes is a good compromise: enough to read the card, not so long that the caller loses patience.

How to decide in a few seconds

The card is designed to be read in a hurry. Look first at the urgency and the reason: if the call can wait, refuse it and call back later at your leisure, using the contact details in the summary. If instead it is urgent or a customer you were expecting, accept and get ready to answer: the phone will ring straight afterwards. If you often do not make it in time, raise the «Maximum caller wait» by a minute, or simplify the instructions so that the conversation is shorter and the notice reaches you sooner.

The module's fields

There are seven voices for the secretary, a subset of those used by «Play a message» and voicemail. Try at least two voices with a real call before choosing: the same text can sound very different.

How to write effective instructions

The «What it must ask or say» field is the heart of the module. The best instructions are short, concrete and well ordered: they say what to ask, in what order and what not to do.

Some rules that work:

• List the questions in the order you want them asked.
• Explain what counts as urgent for you: otherwise the secretary has to guess.
• Tell it what not to do: no prices, no commitments, no promises to call back at a precise time.
• Use your sector's jargon only if your customers use it too.

Example for a law firm

Ask for first name and surname, a phone number and whether the
person is already a client of the firm. Ask briefly for the reason
for the call. It is urgent if it concerns a hearing or a deadline in
the next three days. Do not give legal advice and do not quote fees.

Example for a plumber with an emergency service

Ask for name, phone number and the address of the job. Ask what has
happened. It is urgent if there is a water leak in progress or there
is no hot water in the house. Do not give prices and do not confirm
times: say that the technician will assess the request straight away.

Example for an estate agency

Ask for name, phone number and whether the person wants to buy,
sell or rent. If they are calling about a listing, ask for the
reference code or the street of the property. It is never urgent,
unless the person says they have an appointment today with one of
our agents.

Mistakes in the instructions

The most common mistake is asking too much. Six different questions turn a phone call into a form to fill in, and callers get tired before reaching the end. Keep the essential questions, the ones without which you would not know what to do, and leave the rest for the call back.

The second mistake is leaving room for improvisation. If you do not write that the secretary must not give prices or timescales, it might still try to answer a direct question. Exclusions must be written out clearly, one per line if needed.

The third is not testing. Call your number, pretend to be a customer in a hurry, then a confused one, then one who speaks another language, and read the card you receive. Correct the instructions wherever the summary does not tell you what you need.

To go deeper into writing instructions for voice assistants, there is the guide on instructions for an AI voice agent: the same principles apply here.

Where to put it in the flow

The secretary works best after a first ring step, not instead of ringing. That way the calls you can answer reach you straight away, and the secretary only steps in for the others. A typical flow:

1. «Ring SIP accounts», 20 seconds.
2. «AI secretary».
3. «Ring SIP accounts», 30 seconds, on the same phones or on the app.
4. «Voicemail».

Step 4 is there in case you accept but then cannot answer: the caller can still leave a message. If instead you refuse, the secretary ends the call itself and step 4 is never reached.

Variant: the secretary as the first step

Some people prefer the secretary to always answer first: for example a professional who spends most of the day in meetings and wants to know who is calling before being interrupted. In that case remove the first ring step and put «AI secretary» at the start of the flow. The cost goes up, because every call goes through a conversation with the secretary, but every call reaches you already screened. If you choose this route, keep the opening phrase very short: callers often do not expect to speak to an automated assistant straight away.

All the modules and the rules for chaining them are in the guide to the call routing modules; for the ringing after you accept, see the one on simultaneous or sequential ringing.

Notices and costs

The secretary is charged per minute of conversation, that is for the time it spends talking to the caller, plus any notice SMS. Before turning it on, ask support for the current per-minute rate. The rest of the flow follows the usual costs, described in the price list.

If you use Android, the app notification saves you the cost of the SMS. On iPhone, for now, the notice arrives by SMS. A small tip: well-written instructions make the conversation shorter, so they also cost less.

Limits and privacy

The AI secretary can misunderstand or make mistakes. An unusual name transcribed wrongly, a phone number with one wrong digit, a reason summarised inaccurately: it happens, especially on poor lines or with background noise. That is why the card also contains the transcribed conversation, and why it is always worth asking for contact details and, in important cases, checking them when you call back.

Then there is the question of privacy. Callers are talking to an automated system that collects their data and transcribes the conversation: they must be clearly informed. The simplest way is to say so in the opening phrase, for example «this is the firm's automated assistant». In general, in the EU, Regulation (EU) 2016/679 (GDPR) requires you to inform data subjects about the processing of their data (Article 13), and Regulation (EU) 2024/1689 on artificial intelligence sets transparency obligations for systems that interact with people. How to apply these rules to your case, and what to write in your privacy notice, should be decided with your data protection adviser. If you also record calls, read the guide on call recording.

When you need a full AI voice agent

The AI secretary collects the details and hands you the decision. If instead you want an assistant that actually answers customers, with menus, call queues, opening hours and passing the call to the right agent, you need a PBX: Vohippo does not have these features. The AI voice agent in the virtual PBX from our partner robottino.ai is designed precisely for this.

In that case the roles are split: the PBX handles the conversation and the routing, while Vohippo remains the provider that supplies the numbers and lines, for example with a SIP trunk. To understand the differences and choose, there is the guide on the AI virtual PBX.

Frequently asked questions