Case study · Healthcare

Voice receptionist and patient admin for a specialist practice

A voice receptionist built to answer and book calls, with day-before reminders, a consent capture flow and a morning digest for the practice team.

Engagement facts

Deployment status
Tested prototype. Built and tested, but not in daily use. Built and tested with the practice. The engagement ended before go-live, so it never went into daily use.
Client identification
Client not named. The practice is not named, and its speciality and city are left out because together they would identify it.
Operational context
A specialist medical practice in South Africa whose reception line is busiest while staff are with patients.

The business problem

  • Reception lines are busiest exactly when staff are with patients, so calls ring out or go to voicemail.
  • Reminders, consent forms and the next day's list are assembled by hand.

Why it matters

  • A missed call at a specialist practice is often a patient who books elsewhere.
  • Manual admin pulls skilled staff away from patients.

The solution

  • A voice agent that answered calls, handled bookings and took messages, with clear rules about what it must never say.
  • Day-before SMS reminders, a consent capture flow, a morning digest of the day ahead, and an error alert that tells a person when something fails.

The architecture

  • A hosted conversational voice platform for the agent, connected to n8n workflows for booking and notifications.
  • Separate workflows for voice handling, consent capture, error alerts, SMS reminders and the morning digest, so one failing does not stop the others.
  • Changes to the practice's workflows were locked: nothing was edited without explicit approval.

Integrations

  • Telephony (a phone number connected to the voice platform)
  • SMS for reminders
  • The practice's diary for bookings

Human escalation and exceptions

  • Medical questions are never answered by the agent; the caller is booked or a message is taken for the practice.
  • Failures in any workflow alert a person instead of failing silently.

How it was tested

  • Tested on real calls during the build, reviewing behaviour in the voice platform's call history.

Verified outcomes

  • Tested on real calls during the build, using the voice platform's call history to review behaviour.

    How it was measured: Development and acceptance testing. No production metrics exist because it is not in production use.

Not measured: No operational outcomes are claimed. The system did not go into daily use.

Limitations

  • It never went into production use; the engagement ended before go-live.
  • Testing showed settings that would have needed tuning before any go-live, such as silence handling and call-length caps, so a stuck call cannot run up cost.
  • Warm transfer to a person depends on the telephony integration used.
  • Call recording and retention were never decided, because the engagement ended first.

Confidentiality

  • No patient information, call recordings or practice details are published.

The demonstration

Princess is a scripted prototype with a fictional dental practice. It is not the practice's agent.

Try the Princess prototype

Tell us where the work piles up.

Fifteen minutes on a call or WhatsApp. You describe the job that keeps landing on a person. We tell you honestly whether a system should do it, and what that would take.