AI Phone Assistant in a Medical Practice: Real Results After 9 Months

Be honest: who hasn't tried to call their doctor and simply not gotten through? Probably everyone. This is exactly the problem Dr. Martin Baxmann solved with an AI phone assistant in his practice. And not as a nice experiment, but in the real day-to-day of a practice with five locations.

The results after nine months? Zero missed calls. Over 80% of patients take part. And the whole thing costs €250 a month. Sounds almost too good. But it isn't.

Why an orthodontic practice with five locations switched to AI telephony

Martin Baxmann is an orthodontist in North Rhine-Westphalia. Five practice locations under the name Orthodentix, plenty of patients, and a phone that practically never stops ringing. Around 3,500 calls per month come in across all locations. Sounds manageable? It isn't. On average, 1,000 of those calls went unanswered.

Before, it all ran through an in-house call center with real staff. That basically worked, but with the usual problems:

Martin looked at several all-in-one systems marketed specifically to the healthcare industry. His verdict was pretty clear: "None of them really satisfied me. Everything somehow mixed together, but nothing done really well."

The solution: one AI voice agent for all five locations

Instead of buying a bloated all-in-one package, Martin chose a focused AI phone assistant. And it all happened step by step.

Step 1: Start with after-hours only

In the beginning, the AI voice assistant only picked up when nobody was at their desk anyway. That meant lunch breaks, after closing time and weekends. The assistant did two things:

After every conversation, an automatic email goes to the practice team. With all the info included. The team can then prepare and make the callback in a targeted way.

Step 2: Platform switch and more control

The first version ran on an external provider. That worked, but with limitations. Moving to our own AI phone software then really picked up speed:

Martin's take: "The move went seamlessly, not a single outage. And the platform is intuitive, not overloaded. Everything you need is in there."

The hard numbers: before vs. after

The usage rate is over 80%. So more than four out of five callers actually have the conversation with the assistant. The average call duration? One minute and 13 seconds.

Why 80% of patients go along (and don't hang up)

That was the biggest worry before launch. Especially among the practice staff. Would patients even want to talk to an AI?

The right question isn't: "Human or AI?" but: "Which would you rather have: an AI that picks up right away? A voicemail nobody leaves a message on anyway? Or endless ringing into nothing?"

And that's where it becomes clear. Patients don't compare AI to humans. They compare AI to being unreachable. And the friendly voice assistant wins hands down.

Really cool: patients who run businesses themselves even ask about the solution. Whether it's a home care service, a trade business or a service company.

And one more unexpected bonus: anyone who yells at the AI on the phone would do the same to the real team. So the assistant effectively pre-filters as well. Martin says: "Nobody should have to talk to that person in the first place."

How the AI phone assistant is built technically

No rocket science, promise. The whole assistant is based on a prompt system with a few clever building blocks.

Voice and opening message

Martin chose a natural-sounding female voice. The opening message is deliberately kept short. The caller should get to speak quickly.

Two-path logic: answer or capture

Every call falls into one of two categories:

Automatic summary by email

This is the centerpiece for day-to-day practice work. After every conversation (from 15 seconds in length), a second AI analyzes the transcript and pulls out the most important info: name, phone number, reason for calling, location, summary.

It all lands as a finished email with the practice team. No more listening to voicemails, no more guessing. Just read, call back, done.

Five locations, one assistant

All five practices run through a single assistant. The prompt contains instructions to ask for the preferred location. And if someone speaks unclearly on the phone, the assistant automatically assigns the place to the phonetically closest location.

GDPR and data protection: Is this even allowed?

Short answer: Yes, if you do it right. And Martin did it right.

Calls are not recorded. Only the extracted data (name, reason for calling, contact info) is sent to the team by email. The solution is built to be GDPR compliant.

Still, a tip: anyone who wants to introduce something like this should clear it with their data protection officer beforehand. And inform patients transparently.

What comes next?

The current solution is running, but there's more to come. Three extensions are on the roadmap:

FAQ: The most common questions

Conclusion: AI telephony in a medical practice works. Period.

This case study shows quite clearly: an AI phone assistant in a medical practice is no longer a pipe dream. It works. Today. With measurable results.

Zero missed calls. Over 80% acceptance. €250 a month. And a practice team that finally has time again for patient care instead of the phone.

If you're struggling with reachability, you don't have to overhaul everything at once. Just start outside of office hours. Gather experience. Then expand step by step. That's exactly how Martin did it. And that's exactly how it worked.