Your AI receptionist isn't one piece of software — it's three systems chained together, plus your practice's configuration as the brain. Understanding the pieces makes it easier to debug odd behavior and write content the AI uses well.
The pipeline, end to end
- Caller dials your FrontDesk number. The call hits our voice provider.
- Speech-to-text (STT) transcribes the caller's words into text, streaming as they talk.
- Language model (LLM) reads the transcript plus your practice context (services, hours, FAQs, custom instructions) and decides what to say next.
- Text-to-speech (TTS) turns the LLM's reply into natural-sounding audio using the voice you picked.
- Audio plays back over the phone line. The caller hears the answer.
- Repeat for every back-and-forth (each one is called a "turn").
Total time from the caller finishing a sentence to hearing a response — typically 1–2 seconds.
What's a "turn"?
A turn is one exchange — the caller says something, the AI replies. A typical appointment-booking call is 6 to 12 turns:
Caller: "Hi, I'd like to book a cleaning." AI: "Sure! What day works for you?" Caller: "How about Thursday afternoon?" AI: "I have 2:30 with Dr. Park or 4:00 with Dr. Lee. Which would you prefer?"
Each pair above is one turn.
Where your configuration plugs in
The language model doesn't know anything about your practice unless you tell it. Every call includes:
| What you configure | Where the AI uses it |
|---|---|
| Practice name, hours, address (Settings → Practice) | Greeting, "are you open?" questions, closing |
| Services (Settings → Appointments → Types) | Matching what the caller asks for to bookable slots |
| Greeting (Settings → AI Voice → Greeting) | First thing the caller hears |
| Custom FAQs / Knowledge (Settings → AI Voice → Train) | Answers to common questions ("Do you take Delta Dental?") |
| Persona instructions (Settings → AI Voice → Personality) | Tone, formality, what to call patients |
| Transfer rules (Settings → Phone → Transfers) | When to hand off to a human |
| After-hours mode (Settings → Phone → After Hours) | Behavior outside business hours |
Anything not configured falls back to safe defaults. The AI won't invent a price, a provider, or a service that isn't in your settings — see What your AI can and can't do.
Why responses sometimes feel slow
Most "slow" calls trace to one of these:
- Long answers. A 4-sentence reply takes longer to generate and to speak than a 1-sentence reply. Keep custom instructions concise.
- Network blip. Occasional packet loss between the caller's carrier and our voice provider adds 200–500ms.
- Knowledge-base lookup. If your FAQ is very large, the AI may take a beat to find the right answer.
- Calendar check. Live availability lookups during booking add ~300ms per query.
If most calls feel sluggish, see Troubleshooting → Slow AI responses.