A voice AI agent automating the phone reception desk
A voice AI agent that answers and places calls, checks availability, writes the appointment into the calendar and sends the confirmation — for hotels, clinics and services.
Caller is our own product rather than client work — a voice AI that takes the reception desk on the phone. It speaks naturally, follows the context of the call, and works directly against the company's calendar and email. It is not a phone menu clicking the caller down a tree: Caller asks and answers like a person on the other end, and once it has what it needs, it books the appointment itself. We deploy it where the phone rings all day and every answer is the same handful of questions — hotels and accommodation, medical and dental practices, beauty and wellness, local services and repairs.
Businesses with a reception desk lose calls at two predictable moments: at peak, when staff are serving the customer in front of them, and outside opening hours, when nobody picks up. On top of that sit the repeated questions about availability, prices and opening times, which cost time even though the answer never changes. And then the last step: an appointment agreed on the phone has to be typed into the calendar by hand, which is where double bookings and typos come from. Caller exists to take that whole routine layer — from picking up to writing it down.
Designing a voice agent, we worked on the four things such a system stands or falls on.
- A voice that does not sound like a machine
Callers hang up the moment they recognise a robot. The agent has to hold a natural pace, sound like a person and follow the context of the whole call rather than the last sentence — and the voice can be tuned to a particular brand's tone.
- Booking without double entries
The agent must not offer a slot that is already taken, or write two calls into the same one. It therefore checks availability against the live calendar rather than a copy that refreshes on a timer.
- Knowing when to hand the call over
What matters is not what the agent can do but whether it knows its limit. When a request falls outside its scope it says so, transfers to a live operator or takes a message — instead of inventing an answer.
- Personal data over the phone
At a practice, a caller will often mention their own health — a special category of personal data. Processing had to be built for GDPR from the start rather than bolted on afterwards.
We built Caller to work in both directions — not only as a line that picks up, but as an agent that places its own calls to confirm or remind. Every call ends in a record: the appointment goes into the calendar and a confirmation email goes to the client, without anyone on the team touching a keyboard. Deployment is deliberate — onboarding configures the rules, opening hours and scenarios for the specific business, so the agent talks about what that company actually offers.
- Inbound calls — answered immediately, questions handled and the appointment booked in real time
- Outbound calls — the agent rings clients itself to confirm or remind them of an appointment
- The calendar entry and the confirmation email happen automatically, with no staff involvement
- Integrations with Google Calendar, Outlook, email and CRM; internal systems over an API
- Transfer to a live operator, or a recorded message, whenever a request falls outside scope
- Voice tunable to the brand's tone and style, with GDPR-compliant data processing
Mapping the calls
We go through what people actually ring about, and which call types repeat often enough for the agent to handle unaided.
Rules and scenarios
We configure opening hours, the offering, the agent's limits and the situations in which a call passes to a person.
Calendar and email
We connect the agent to the company's calendar and email, and to an internal booking system over an API where there is one.
Testing and go-live
We run real calls, tune the wording and the voice, and only then switch the line over to the agent.
Caller moves the reception desk from opening hours to continuous cover: a call gets handled while staff are with a customer, and after closing. The routine layer — availability, booking, repeated questions, confirmations — runs unaided, and whatever the agent cannot handle is not lost but passed to a person with a record of the call. The appointment lands straight in the calendar, which removes the step where a phone conversation is retyped by hand and double bookings appear. What is left for the team is the customer standing in front of them.