Where it fits
For a plumber, a clinic or a gym, most missed calls are simple: hours, location, "do you do X", and "can I book". A voice agent handles those well, twenty-four hours a day, in a natural voice, and it books the appointment into the GoHighLevel calendar while the caller is still on the line. The alternative is voicemail, which most people no longer leave. Against that baseline, voice AI is a clear win for after-hours and overflow.
Where it doesn't yet
- Complex scheduling. "Can I have Tuesday but only if Dr. Collins is in and it's after school?" still trips it up.
- Upset callers. An angry customer needs a human immediately, not a calm synthetic voice.
- Bad lines and accents. Transcription errors compound. A noisy van on speaker is a hard environment.
- Judgment calls. Anything involving price negotiation, medical advice or an emergency.
For these, the agent should recognise its limits early and transfer or promise a call-back, rather than persist.
Designing the call
We script the call like a short form: greet, ask what the caller needs, ask one or two qualifying questions, offer two slots, confirm, and say what happens next. The agent has the same knowledge base and rules as the text bot from the prompt template. Hand-off triggers include any sign of frustration, a request for a person, and two failed attempts to understand.
The agent introduces itself as the practice's virtual assistant. Callers are fine with that when it is helpful and honest. They are not fine with discovering it later.
Voice plus text-back
Voice never runs alone. The moment a call is missed, missed-call text-back fires as well, so the caller has a written channel even if the voice agent struggles. Anything the agent captures is written to the contact record, and if it hands off, the text thread already exists for a human to continue on.
Measuring it
Three numbers: percentage of after-hours calls answered, percentage of those that end in a booking, and hand-off rate. Someone listens to a sample of recordings every week for the first month. A good deployment books a third or more of eligible calls and hands off around a quarter. If callers hang up in the first fifteen seconds, the greeting is too long or too robotic; shorten it.
Key takeaways
- Voice AI is good at after-hours intake, FAQs and booking simple appointments.
- It struggles with complex scheduling, upset callers, poor lines and anything needing judgment.
- Design the call as a short script with a booking outcome, and let it hand off gracefully.
- Run text-back in parallel so the caller has a channel that never fails.
- Listen to recordings weekly and track booking rate and hand-off rate.