What happens during the call, second by second
When a customer dials your number, the call is forwarded to the agent, either because you have set up call forwarding from your current line or because you have chosen a dedicated number. From there, five things happen in a loop, with every sentence:
- It listens. The caller's voice is transcribed into text as they speak, even with an accent, background noise or a hesitation.
- It understands. A language model reads that transcription in the context of your business: your opening hours, your services, your rules and what has already been said during the call. From this it works out the intent: book an appointment, report a breakdown, ask a question.
- It decides what comes next. Depending on the call flow you have defined, it answers, asks a follow-up question, records a piece of information or triggers a transfer.
- It speaks. The reply is converted to speech and played to the caller. The whole thing takes under two seconds, which keeps the rhythm of a normal conversation.
- It stops talking when the caller speaks. The agent detects interruptions and picks up where it left off.
None of this is visible to the caller. They feel they are talking to someone who knows the business well, who says clearly that they are a virtual assistant, and who knows where to stop.
Where its answers come from
This is the question that matters most, and the one people ask least. A generic voice agent answers from what it has learned on the internet, which makes it chatty and imprecise. A well-designed AI receptionist only answers from three sources:
- Your description. What your business does, where, when and for whom. With Driing, this is the starting point: you describe your business and AI generates a first agent.
- Your approved answers. The prices you are willing to quote, the late check-in procedure, the list of documents to bring. Everything written in the editor is the reference.
- The call flow. The order of the questions, the information to collect, the cases that trigger a transfer.
When a question falls outside that frame, the agent says so, offers to pass the question on and notes the caller's details. It doesn't make things up. This rule is not a detail: it is what makes the tool usable in a medical practice or a law firm.
The call flow: the difference between a chatbot and a useful agent
A chatbot waits for a question and answers it. A receptionist works: it greets, identifies the reason for the call, asks the necessary questions, acts, then wraps up. This flow is built visually, like a diagram:
Greeting in the garage's name → Reason for the call? → Appointment: vehicle, number plate, service, slot → Confirmation. Fault: description, mileage, call-back. Roadside breakdown: transfer to the workshop. Other: message and contact details.
Each step has its own instructions, in your words. You can add steps, remove them, change a phrasing and test the change by calling the agent. No line of code is needed: that is the principle of the visual editor.
Acting: note, transfer, book
Answering is not enough. What your customers expect is for their request to go somewhere. Three actions cover almost every case:
- Note. The agent collects the information you have defined for each type of request and always structures it the same way.
- Transfer. For the situations you have classed as urgent, it puts the call through to the number of your choice, day or night. See the use case Emergencies and transfers.
- Book an appointment. It checks your diary, offers slots that are genuinely free and books the appointment there. See Appointment booking.
- Look things up and act in your tools. Depending on your set-up, the agent reads your availability, finds a client record or checks a detail in your business software, then acts: it books, moves, cancels, records. You decide what it can read and change.
After the call
Every conversation produces a written summary for your team: who called, why, what was answered, what remains to be done and how urgent it is. This summary is what saves time, far more than the call itself: no more inaudible messages, no more calling back blind. An audit log also keeps a record of every call handled and every change made to the agent.
And the data?
A call contains personal data: a name, a number, sometimes a piece of health information or a dispute. Three questions to ask any provider: where is the data hosted, who has access to it, and how long is it kept. With Driing, the data is hosted in Switzerland, the agent always announces that it is a virtual assistant, and you review the retention and deletion rules before going live. The details are on the Security and data page, and the article on the FADP explains what the law requires.
What it doesn't do
An AI receptionist doesn't diagnose, doesn't give legal advice, doesn't negotiate and doesn't decide on an exception. It collects, answers within the frame you have set and hands over. That limit is precisely what makes it reliable. We devote an article to it: what voice AI can do, and what it must not do.