A transparent evaluation methodology for AI phone assistants in DACH-market medical practices: 7 dimensions, 28 criteria, weighted scoring logic. No fake rankings.
bhomy Team
May 3, 2026
14 min read
Scroll
What this document is — and what it isn't
This is a methodology, not a market report. We don't publish ratings of third-party vendors that we haven't tested ourselves in a controlled practice pilot. Instead we provide an evaluation framework that practices, IT leads and medical-care-centre (MVZ) management in the DACH market can use to structure vendor selection. bhomy's own scores appear at the end — with a date, a methodology note and traceable reasoning.
7
evaluation dimensions
28
individual criteria
0
fabricated comparison ratings
In 2026, AI phone assistants for medical practices are a market with more than 20 visible vendors across the DACH region (Germany, Austria, Switzerland) — from German-Swiss specialists through US platforms with German localisation to regional white-label solutions. A surface check (website, demos, sales calls) doesn't get a practice to a sound decision. We propose a consistent evaluation framework.
01
Audience and scope
The methodology applies to GP and specialist single and group practices, as well as medical-care centres (MVZ), in Germany, Austria and German-speaking Switzerland, with a call volume between 60 and 250 calls per working day. Larger structures (hospital outpatient clinics, practice chains) require extended dimensions (multi-tenant, SLA hierarchies) not covered here.
We consider only AI voice assistants that conduct phone conversations autonomously. Pure IVR systems ("press 1 for …") and answering machines with transcription are deliberately not placed in the same framework — they solve a different problem.
02
The seven evaluation dimensions
1. Data protection and legal compliance (weighting 25%)
01EU/CH data residency: contractually guaranteed, documented, not "best-effort".
02A data-processing agreement under Art. 28 GDPR (DE/AT) or under Switzerland's revised Data Protection Act (revDSG) — formally complete, not just "available on request".
03Assumption of professional confidentiality (§ 203 of the German Criminal Code; § 54 of the Austrian Medical Act; Art. 321 of the Swiss Criminal Code).
04Retention: audio recordings disabled by default, transcribed summaries with a configurable retention period.
This dimension has the highest weighting because non-compliance disqualifies the whole solution. A practice must not work with a vendor that won't contractually accept medical confidentiality — no matter how convincing the demo was.
01DE: the common practice systems (e.g. Medatixx, T2med, CGM, RED Medical).
02AT: the common Austrian systems (e.g. InnoMed, RZA).
03CH: the common Swiss systems (e.g. Vitomed, Aeskulap, Triamed).
04Method: a direct API/REST integration preferred, healthcare messaging/HL7 acceptable, an email task only as a backup.
A practice whose system isn't supported has manual handover paths — operationally a step back, not forward. The methodology assesses concretely: which systems does the vendor write to directly? Which only as an email task? Which not at all?
3. Speech quality and latency (weighting 15%)
01STT accuracy on medical terms (diagnoses, medications, ICD codes) — a word error rate below 8%.
02TTS naturalness: rated by reception staff in a 30-call sample (scale 1–5, threshold ≥ 4).
03End-to-end latency below 800 ms (95th percentile) between the end of speech and the start of the assistant's reply.
04Barge-in works robustly (the patient can interrupt the assistant without the call crashing).
4. Medical triage safety (weighting 12%)
01Detection of acute symptoms (chest pain, shortness of breath, loss of consciousness, severe bleeding) and immediate referral to the emergency number (112 in the EU; 144 in Switzerland).
02No independent medical assessment ("you probably have …" is forbidden).
03A clear handover to staff for medical follow-up questions.
04Documented escalation paths at night and on weekends.
Triage safety is non-negotiable
A vendor whose assistant gives even one independent medical assessment in a test sample is disqualified for use in a medical practice — regardless of its score in other dimensions. That's medical-professional law, not a UX preference.
5. Appointment-booking logic (weighting 10%)
01Avoiding double bookings on parallel calls (a concurrent-call lock).
03Correct treatment-duration estimation (the practice's consultation grid).
04Cancellation and rescheduling without staff intervention.
6. Operating economics (weighting 10%)
01A transparent price structure (€/month or €/call minute, not opaque hybrid models).
02Predictable scaling as call volume grows.
03No setup fee above €1,500 net for standard practices.
04A minimum term of 12 months or less.
7. Support, onboarding, ownership (weighting 10%)
01Local-language support within usual response times (first response in ≤ 4 working hours).
02Onboarding support with clinical/staff training (at least 2 hours, in person or by video).
03The vendor's ownership: who holds the majority? Where is it based? Are there investors from third countries with a problematic data-protection adequacy situation?
04A transparent exit: how does the practice get its data back when switching vendor?
03
Scoring logic
Each dimension is scored on a scale of 0 to 5 (0 = not met, 5 = excellently met). The overall score is the weighted sum and produces a value between 0 and 5.0.
01Score ≥ 4.2 = recommended for use in the tested practice constellation.
02Score 3.5–4.1 = fundamentally viable, with documented limitations.
03Score 2.5–3.4 = usable only in special cases; risks outweigh standard benefits.
04Score < 2.5 = not recommended for regulated medical professions.
Knock-out criteria override any score
Three criteria are knock-outs: (1) no EU/CH data residency, (2) no assumption of medical confidentiality, (3) independent medical assessments in the test. Fail any of these and you're disqualified — regardless of the weighted overall mark.
04
bhomy self-assessment (as of May 2026)
We apply the methodology consistently to ourselves too. The following assessment is self-transparency, not advertising — the weaknesses are published as well.
4.5 / 5
data protection & law (EU hosting, DPA, medical confidentiality)
4.0 / 5
PVS/PIS integration (8 systems direct, 2 as an email task)
4.3 / 5
speech quality & latency
4.6 / 5
medical triage safety
01Strength: full GDPR/revDSG compliance including medical confidentiality in the contract.
02Strength: triage logic with hard safety rules (acute symptoms → emergency number) and no independent assessment.
03Weakness: three rarer Austrian and Swiss practice systems are currently connected only via an email task, not directly.
04Weakness: premium TTS voices in Swiss German are not currently in the standard scope.
Why we don't score competing vendors
We considered rating third-party vendors several times. We refrain from doing so as long as we can't set up pilot practices with an identical test constellation for each vendor — anything else would be methodologically dishonest and would strip this document of its value. If you run a practice and would like to take part in an independent, controlled vendor comparison, write to us.
05
How to use this methodology in your practice
01Define 30 typical call reasons for your practice (appointment, repeat prescription, sick note, acute, private billing, e-prescription query etc.).
02Ask each vendor for a pilot phase of at least 14 working days with your real call reality — not demo calls.
03Rate each of the 7 dimensions based on the pilot output (call logs, staff feedback, PVS integration tests).
04Multiply each rating by the weighting and sum it up.
05Check the knock-out criteria — if one disqualifies, the weighted mark is irrelevant.
06Document the assessment in writing. When switching in 12–24 months, that documentation is your most important basis.
06
Versioning and updates
This methodology is reviewed at least annually. Changes are documented in the update log at the end. As of May 2026, the regulatory anchors (GDPR, revDSG, the EU AI Act as of Q2 2026, and the relevant national medical-association guidance) are current.
A practice working with a non-GDPR-compliant vendor risks fines of up to 4% of global annual turnover, plus civil damages claims from patients. Operationally that isn't a nice-to-have but an existential risk. A 25% weighting reflects that appropriately — anything lower trivialises it.
Request a pilot phase using this methodology
If you run a practice and would like to evaluate bhomy in a controlled pilot against exactly these criteria — we're happy to set that up. Not a sales demo, but a real 14-working-day pilot with your call reality.
We use cookies to provide you with the best possible experience on our website. Some of them are technically necessary, others help us improve the website. Read Privacy Policy
