An AI voice agent with one clear job
We start with one type of call: booking an appointment, taking a meter reading or logging a request when you are busy or outside opening hours. The agent introduces itself as an AI system. We check that it understands the request, records the outcome in the calendar or business software you use, and hands over when a person is needed.
What is an AI phone assistant?
It is an AI system that answers calls, interprets what callers say and carries out a defined task, such as booking an appointment, logging a request or collecting information. Callers speak to it instead of navigating a phone menu. Its scope is set in advance, including when it must hand over to a person.
The problems to solve
Calls interrupt the work you are doing
The phone rings while you are on site, in the workshop or with a customer. You rush to answer or cannot pick up, and the request may never get recorded.
The same questions, all day
Opening hours, progress on a request, available appointment times: short, recurring enquiries that keep pulling you away from your work.
Out-of-hours calls go unanswered
Someone calling in the evening or on a Saturday may get no answer. If they leave no message, you have no record of what they needed and nothing to follow up.
What we build
One clearly defined task
We work with you to choose the type of call the agent will handle: a booking, a meter reading or collecting specific details. We define what it can complete and which requests must go to a person.
A clear handover to a person
Callers should not have to keep asking to speak to a person. We agree who takes transferred calls and how to handle the request if no one is available. The person taking over also receives the details already collected.
Bookings in the calendar you use
We check that bookings are saved in your calendar and requests in your business software, with the right details. A spoken confirmation alone does not mean the task is complete.
Testing with realistic calls
Before it goes live, we test it on the kinds of calls you receive, including accents, background noise, unfinished sentences and unrelated requests. We check what happens when it misunderstands, which details it records and how it hands the call over to a person.
What we do not do
We do not replace reception or customer support
The agent handles tasks that can be completed over the phone in a few steps. Complaints and support requests stay with people. Callers who ask to speak to a person must not be forced to continue with the agent.
We do not make cold calls
We do not build systems that work through contact lists making sales calls. The agent answers incoming calls or calls back someone who left their details expecting a reply.
We do not pass it off as a person
The agent identifies itself as an AI system at the start of the call. Callers can then decide whether to continue or ask to speak to a person.
The work your team still needs to do
Handing some calls to an agent creates ongoing work: reviewing calls, identifying misunderstandings and correcting the system. Someone needs to be responsible for these checks and have time set aside for them. That work continues after launch.
Reviewing calls that went wrong
Dropped calls, misunderstood requests and handovers need reviewing to identify what needs fixing. Reviews are more frequent at the start; later, we assess how to continue, including whether to sample calls. Reviewing remains an ongoing responsibility.
Being there to take over
When the agent transfers a call, someone has to answer. You need to know who will do that and when they will be available. If the transfer ends in voicemail, the customer is still waiting for help.
Checking saved appointments
Check each booking at the start: it needs to be in the right calendar, with the correct date, time and customer details. Finding and fixing mistakes also takes time that must be accounted for.
What to measure in the trial
- How many calls do you get in a week, and how many involve the same recurring task?
- How many calls go unanswered, and at what times of day?
- Out of a hundred calls handled by the agent, how many are completed correctly, with the details saved where they belong and no human intervention?
- How often do callers become frustrated or ask for a person, and how long do they wait to speak to one?
- How many hours a month are needed to review calls and correct misunderstandings or booking errors?
The facts and their sources
Article 50 of the AI Act has applied since 2 August 2026: people interacting directly with an AI system must be informed of it, unless this is obvious. On the phone it is not obvious, so the agent says so at the start of the call.
Questions we often hear
- Will customers accept it?
- We do not assume they will. Anonymous accounts from suppliers in Italy are mixed: one reports withdrawing a voice agent for pizzerias because it was poorly received, while another reports making sales. These accounts do not establish that the Italian market is ready. We test how callers respond to the specific task; complaints and customer support stay with people.
- What happens when it does not understand?
- It asks for clarification once. If that does not resolve the problem, it starts the handover to a person. If no one is available, it must say so; we agree in advance how to capture the request. Misunderstood phrases need reviewing so the system can be corrected.
- Will bookings appear in the calendar you use?
- We first check that the agent can connect to the calendar you use. Then we test the whole process, from the call to a saved appointment with the correct details. The booking is complete only when it appears where you need it.
- Can it call customers?
- It can call back people who left their details expecting a reply. We do not build systems that work through contact lists making sales calls.
- Are the calls recorded?
- We agree with you whether to record calls and how to inform callers. When reviewing an error, we assess whether a transcript of the relevant exchange is enough.
- How does it handle accents, dialects and background noise?
- We test it against the kinds of requests you receive, including accents, dialects, background noise and unfinished sentences. We examine mistakes and check the handover to a person. If it cannot handle the chosen task reliably, we will tell you.
- What does it cost to keep running?
- You need to account for call charges and the time spent reviewing calls, correcting the system and checking bookings. We estimate these with you before starting and check them during the trial. Time for supervision must remain part of the ongoing workload.
Start with one kind of call
Tell us about a request you get often: what information is needed, where it should be saved and when a person needs to step in. We use that to assess whether a voice agent makes sense and plan a trial with realistic calls.