A phone line that always answers

Most organisations lose money on the phone in a way that never shows up in any report. The calls that ring out at 11am, the voicemails nobody returns, the person who tries once, gets nothing, and books with somebody else. You cannot see any of it, because a call that does not happen leaves no trace.

An AI receptionist answers every one of them. It picks up on the first ring at three in the morning, it never puts anyone on hold, and it does not have a bad day. What it should not do is pretend to be a person or try to handle things it has no business handling, and most of the work in building one well goes into that second part.

Who this is for

• Clinics and practices where reception is one person who is also doing three other jobs, and where the phone competes with the patient standing at the desk.
• Trades and services (installers, garages, cleaning firms) where the people who could answer are on a job, and where a missed call is simply a lost customer.
• Any business with a predictable twenty questions: opening hours, parking, what to bring, do you take this insurance, how much is a first appointment.

What it actually does

The agent answers, identifies why someone is calling, and then does one of four things. It answers the question outright, from a set of facts you control. It books, moves or cancels an appointment directly in your calendar. It takes a message and routes it to the right person. Or it transfers to a human, immediately, because the call is one you would want a human to take.

That last path is the one that matters. We agree the handover rules with you before writing anything: which topics always go to a person, what happens outside working hours, what the agent does when someone is upset, and what it says when it simply does not know. An agent that cannot gracefully admit defeat will eventually embarrass you in front of a customer.

Everything the agent says about your business comes from a source you can edit. When your prices change or you close for a week in August, you change one thing, not a prompt buried in someone else's platform.

What a build looks like

• A phone number, or your existing one forwarded, with the old one still working throughout.
• The voice agent itself: greeting, disclosure, intent handling, the twenty questions, and the escalation rules.
• Calendar integration, including how it handles double bookings and buffer time.
• A transcript and recording trail, so you can hear what it actually said, and correct it.
• A staff handover process, because the humans need to know what the agent has already told someone before they pick up.
• Dutch, English, or both.

We run it in parallel with your existing setup first (taking overflow calls, or evenings only) so you can hear it working on real calls before it becomes the front door.

Where the data goes

Call audio and transcripts are personal data, and if you are a clinic they may be health data, which carries a much higher bar. We design for that from the start: European processing where the data calls for it, retention that you set rather than inherit, access limited to the people who need it, and a full record of who listened to what. If your callers are patients, we will also tell you where the line sits between an assistant and something a regulator would treat as a medical device, and keep you well clear of it.

What shapes the cost

We do not publish a price for this, because the range is genuinely wide and any number we printed would be wrong for half the people reading it. What moves it:

• How many things it needs to do. Answering questions and taking messages is a short build. Booking into a live calendar with real availability rules is a longer one.
• What it has to talk to. A modern system with a clean API is straightforward. A practice system from 2009 with no API is a different project.
• Languages. One is simple. Two means twice the tuning and twice the testing.
• Whether patient data is involved. Health data means additional design, documentation and review, and that is real work rather than a line item.
• Call volume, which drives the per-minute running cost rather than the build.

Tell us how you handle calls today and roughly how many you get, and we will come back with a written scope, a fixed price and a date. That costs nothing and you keep it either way.

Frequently asked questions

Will callers know they're talking to an AI? Yes, and by law they have to. From August 2026 the EU AI Act requires anyone deploying a system that interacts directly with people to make that clear unless it is already obvious. We build the disclosure into the greeting rather than bolting it on, and in our experience it costs nothing: people mind being misled far more than they mind talking to a machine that is quick and accurate.

What happens when it doesn't understand someone? It hands over. Every agent we build has explicit handover rules: an unclear caller, an angry one, anything clinical or financial, or simply someone asking for a person. The failure mode we design for is a warm transfer or a callback request, never a loop. An agent that cannot say 'let me get someone' is a liability, not a feature.

Can it work in Dutch and English? Yes, and it can switch mid-call. Dutch voice quality is now good enough for production, though it still pays to tune the pronunciation of your own product names, street names and staff names, which we do as part of the build.

Does it integrate with the calendar we already use? That is usually the whole point. Google Calendar, Outlook, and most practice management systems with an API. Where a system has no API we look at what it can import or export before recommending you replace anything.

What does it cost to run? Voice conversation is billed per minute by the underlying providers, typically somewhere between a few cents and around twenty cents depending on the models and voice quality you choose. That sits on top of the build, and we size it against your actual call volume during scoping so there are no surprises.