The phone rings at half past eight, and nobody is at the desk.
This is a simulated AI receptionist for a plumbing and heating company. Take the call yourself: choose what the caller says and watch what the agent writes down, what it refuses to guess, and when it decides the call needs a human.
SimulationNothing here calls a phone network or a language model, and no call is placed. The scripted answers are fixed, which is the point: they show the behaviour, the guardrails and the escalation rules. The production build runs the same script against a live speech pipeline, and each outcome says what it would do differently.
Which call is coming inA regular caller booking the annual service.
Playback
Incoming callBooking line, 01423 880 214
- Turns
- 0
- Agent time
- 0.0s
- Fields
- 0
The call
The line is quietAnswer the call to hear the agent open, then choose your lines as the caller.
What the demo is showing
The behaviour
- Opens with the business name and the line, so the caller knows who answered.
- Confirms the address before offering a slot, and updates it when the caller has moved.
- Records each field as it is said, and replaces rather than duplicates.
- Answers in under a second, which is the difference between a caller waiting and a caller hanging up.
The guardrails
- It will not price a bathroom refit on the phone, and it says why rather than guessing.
- It asks about injuries and water near electrics before it dispatches, because that changes who can attend.
- An electrical risk escalates to the engineer and the duty manager together.
- It stays on the line while the caller looks for the stopcock, instead of hanging up.