How a General Practitioner Can Reduce No-Shows by 35% with an AI Appointment Assistant

In a general practice, the phone never stops. Between requests for test results, prescription renewals, and appointment bookings, the medical secretary can spend up to 60% of their time on incoming calls alone. And despite this effort, no-shows persist: slots released too late, forgotten reminders, patients who hang up without reaching anyone.

The good news: a properly configured AI voice agent can change things within a few weeks. Here is how.

The Real Problem: No-Shows That Cost Real Money

According to McKinsey (2024), administrative tasks account for up to 34% of healthcare professionals' time in private practice. A large part of this burden falls on appointment management: confirmations, last-minute cancellations, follow-up calls.

The average no-show rate in general practice in France runs between 15 and 25% depending on the practice and region. On a day with 25 consultations, that means 4 to 6 lost slots, amounting to 120 to 180 euros in evaporated revenue every single day.

According to Gartner (2025), 45% of healthcare organizations plan to deploy conversational agents for appointment management by 2027. This shift is no longer limited to large hospital groups: consumer-grade solutions have now reached a maturity level accessible to independent practices.

The Case of Dr. Sophie Marchand in Lyon

Dr. Sophie Marchand has practiced general medicine at a group practice in Lyon for 12 years. With two associates and a part-time secretary, the practice handled around 80 calls per day. The secretary spent her mornings picking up, hanging up, taking notes, and calling back.

In January 2026, Sophie connected a Vapi voice agent to her Doctolib account via webhook. The protocol was straightforward:

1. The agent answers incoming calls 24/7
2. It reads available slots in real time from Doctolib
3. It confirms or modifies appointments using speech synthesis
4. It sends an SMS confirmation and a reminder 24 hours before
5. If a request falls outside its scope (urgent symptom, prescription), it transfers the call to the secretary or offers a callback message

Result after 10 weeks: 35% reduction in no-shows, 2.5 hours freed up daily for the secretary, and zero medical data processed by the AI (exchanges remained strictly limited to scheduling logistics).

How It Works Technically

1. Choose Your Voice Agent

Two solutions currently dominate for this type of use case:

• Vapi: flexible voice API, compatible with many integrations, pay-as-you-go pricing. Ideal for practices that want to stay in control of configuration.
• The latest versions of ElevenLabs: extremely natural voices, low latency. Preferred when voice quality is the top priority.

For a medical practice, Vapi is often preferred for its ability to disable persistent logs, a condition required for GDPR healthcare compliance.

2. Connect Doctolib via Webhook

Doctolib provides outbound webhooks (event notifications) and a partner API for certain configurations. The webhook allows the AI agent to:

• Query available slots in real time
• Book or cancel an appointment
• Trigger an SMS reminder through the existing Doctolib interface

This connection does not require replacing Doctolib: it plugs into it.

3. Implement the Anti-Hallucination Protocol

This is the most important point for compliance. A poorly configured AI agent can fabricate information: "Your next appointment is on the 15th," when no slot has been confirmed.

Rules to apply:

• No medical responses: if a patient mentions symptoms, the agent replies "I am not authorized to advise you on this, let me transfer you to the secretary."
• Systematic confirmation: the agent always repeats appointment details before ending the call.
• No health data in the prompt: the context provided to the AI is limited to available slots and the practice's identity.
• Immediate human transfer on any sensitive keyword (emergency, pain, treatment, medication...).

4. HDS Hosting and Compliance

To comply with healthcare GDPR and CNIL recommendations, infrastructure must be hosted by an HDS-certified provider (Hébergeur de Données de Santé). In practice, this means:

• Voice recordings (if any) must be stored in France on certified infrastructure
• The AI itself must not receive identifiable health data in its API calls

Bpifrance (2023) notes that 68% of medical practices with fewer than 5 practitioners have no dedicated resource for evaluating digital compliance. Getting support for this step is not optional.

5 Steps to Get Started

Step 1: Map your call flows For one week, log call volume by time slot, type of request, and pickup rate. This baseline will help you size the solution and measure ROI afterward.

Step 2: Choose your tool and provider Vapi, ElevenLabs, or a packaged solution like Synthflow. Check compatibility with Doctolib and the ability to disable logging.

Step 3: Write the agent script The script defines exactly what the agent can and cannot do. Be very precise about human transfer cases. Test on yourself before testing on patients.

Step 4: Connect and test in silent mode Run the agent in parallel with your secretary for 1 to 2 weeks. Compare the AI's confirmations against the secretary's. Adjust edge cases.

Step 5: Go live and measure Start the agent on off-hours slots first (evenings, weekends). Track the no-show rate week by week. Expand gradually.

What This Changes in Concrete Terms

For a 3-doctor practice with 300 consultations per week:

• No-show rate drops from 20% to 13% = 21 recovered slots per week
• At 30 euros per consultation = 630 euros in additional revenue per week
• That is roughly 2,500 euros per month, at a solution cost below 200 euros per month

Return on investment is typically reached within 4 to 6 weeks.

Your Next Step

Do you run a practice and want to assess whether this solution fits your situation? The IMPACT methodology I apply in my assignments always starts with an honest Diagnostic: is the problem real, is the solution proportionate, and do you have the resources to do it properly?

If you want to frame this project without getting lost in technical details, reach out directly. This is also what the TransformAudit is designed for: a 48-hour AI audit to identify the 3 concrete high-ROI actions in your practice or organization.

---

Tarek Nachnouchi is an AI transformation consultant for SMEs and independent professionals. He operates across France from Bordeaux.