For medical practices
The phone is the bottleneck. The law is not.
44 percent of statutorily insured patients are dissatisfied with their practice's phone reachability, while 68 percent of medical assistants spend over 40 percent of their time on organization and administration instead of patients (GKV umbrella association 2024, Fresenius 2024). Since December 2025 the KBV itself classifies AI phone assistants as a low-risk application. The bottleneck is fixable, inside clear legal lines.
30 minutes, no obligation. German or English.
44%
of statutorily insured patients are dissatisfied with their practice's phone reachability
GKV umbrella association patient survey 2024
68%
of medical assistants spend over 40 percent of their working time on organization and administration
Hochschule Fresenius 2024
57.8%
of physicians now do medical-assistant tasks themselves
KV Bremen/Zi 2023
The gap
Your software market does documentation. The phone and the paperwork between systems stay yours.
Doctolib ships a phone assistant and an AI practice system for three specialties, CGM a system-agnostic phone assistant, medatixx an ambient documentation with EU hosting. The built-in AI concentrates on documentation, often as a paid add-on per specialty. What stays open is the day between the systems: phone, fax, secure mail and the patient record, where those 68 percent of assistant time disappear. That is what we build.
Phone intake inside the KBV lines
Appointments booked, moved and cancelled around the clock, with an AI announcement from the first sentence and emergency escalation to a person, always. The KBV classifies exactly this as a low-risk application; evaluating symptoms is out of scope by design.
Prescription, referral and callback requests as structured messages
Requests are captured with the details your team needs, assigned to the patient and queued as a worklist. Your staff work through a tidy list instead of a full voicemail box, and no machine answers anything medical.
Sorting between fax, secure mail and the record
Incoming documents are classified, assigned to the right patient and forwarded to the right person. 77 percent of practices still communicate with hospitals mostly on paper (PraxisBarometer 2025); the sorting and forwarding is exactly where assistant hours vanish.
Patient-record workflows that actually run
The electronic patient record (ePA) has been mandatory since October 2025, yet 81 percent of practices upload documents while only 11 percent have ever downloaded one (PraxisBarometer 2025). We structure filing and retrieval so the record starts working for your practice too.
Health law first
Built inside the KBV lines, not around them
Medical practices cannot use AI the way other businesses do, and any provider who does not open with that is a risk. Our setup follows the KBV's AI guidance of December 2025 and the statutes it points to.
The KBV framework, followed to the letter
The KBV classifies AI phone assistants as low risk, requires an AI announcement at the start of the call, and states that patient consent is generally not required as long as the data is not used for training. Responsibility stays with the physician; the AI makes no decisions.
Confidentiality under § 203 StGB
A data processing agreement plus a formal written confidentiality commitment of the provider, signed before any access to data. Without it the physician is personally criminally liable; with us it is the first document on the table.
§ 393 SGB V and the C5 attestation
Cloud processing of health data is only permitted with processing in the EU, EEA or Switzerland, a domestic establishment of the provider, and a BSI C5 attestation, since July 2025 as Type 2. A phone assistant piping health data through a US AI API without C5 is out, however good the demo sounds.
A hard stop before the medical-device line
The moment software evaluates symptoms, grades urgency or recommends action, it becomes a class IIa medical device under the MDR and a high-risk system under the EU AI Act. Our assistants capture, structure and forward. A human does the grading, in every setup we build.
How it starts
Same engagement, firm-specific scope
Intro call
Free
30 minutes on your firm's bottlenecks. About a third of these calls end with the advice that AI is not the right tool yet.
Process audit
from €2,500 · credited to the implementation
We map where the hours actually go, from intake to deadline board, and rank the use cases by payback. You get a shortlist with expected effort, not a slide deck.
Implementation
Fixed fee, named in your proposal
One process at a time, live in your daily business in 2 to 4 weeks, with your team trained and a sign-off step wherever client data leaves a draft.
Fair questions
What medical practices ask us first
Is AI on the patient phone even legal?
Yes, under conditions, and since December 2025 that is official: the KBV classifies AI phone assistants as a low-risk application, requires an AI announcement, and states that patient consent is generally not needed as long as data is not used for training. The conditions with teeth are § 203 StGB confidentiality and § 393 SGB V with its C5 attestation, and our setup is built around exactly those.
Does this replace our medical assistants?
No, and that is not the goal. In a Bremen survey one practice in four could not fill a single assistant position in twelve months, and 68 percent of assistant time goes to organization and administration. Automation gives part of that time back to the front desk and the patients.
What may the assistant never do?
Evaluate symptoms, grade urgency or give recommendations for action. That is where software becomes a class IIa medical device under the MDR and a high-risk system under the EU AI Act. We deliberately build only on the legal side of that line and say so to every prospect in the first call.
We use Doctolib, CGM or medatixx. Does this replace it?
No. Your practice software stays the system of record, and where its built-in features cover your case, start there; we say so in intro calls. We build the layer between the systems: fax, secure mail and record sorting, callback lists, request capture. Where interfaces exist, we use them.
What does it cost?
The audit starts at €2,500 and is credited in full against the implementation. Implementations are fixed-fee. For market context, we published what AI automation actually costs, with sources, on our blog. Read the article with all market prices.
Talk to us before the next busy signal does
A 30-minute call is enough to tell whether your practice has an automation case. If it does not, we say so.