The AI assistant market for healthcare is growing fast. That means more options and a lot of noise.
The global AI voice agents in healthcare market hit $472M in 2025 (Grand View Research) and is on track to pass $650M in 2026. More vendors, more pitch decks, more variation in what’s actually delivered.
The real question isn’t whether to adopt an AI assistant. It’s how to pick the right one. A poorly configured assistant misses appointments, frustrates patients and creates legal headaches around voice data. Switching providers mid-contract has costs nobody talks about: weeks of migration, reconfigured calendars, patients left unattended during the transition.
These are the 7 criteria worth checking before you sign anything. They come from what works (and what doesn’t) in chiropractic, physio and osteopathy clinics that already run AI assistants.
1. Response latency: 500 milliseconds is the threshold
Latency is the time between when the patient stops speaking and when the assistant starts replying. Above 500ms, the conversation stops feeling natural and the patient starts wondering whether the call dropped.
A clinical voice assistant should sit below 500ms with over 95% accuracy on medical terminology. That includes understanding treatment names, specialties and the colloquial way patients describe symptoms (“my back went out”, “I have a stabbing pain in my neck”).
What to ask the vendor: average latency in production (not in demos), whether servers are in Europe, and how they handle Monday-morning traffic spikes.
2. Integrations with your clinic software
An assistant that doesn’t talk to your management software is a voicemail with a nicer voice. Integration with your calendar is what makes real-time booking possible, avoids double bookings and lets the system check availability without a human in the loop.
Three integration types you’ll see: direct API (most reliable), webhooks (for events like “appointment created” or “cancelled”), and calendar sync (for software without an open API, like Google Calendar).
If you’re starting from scratch and want the cleanest setup, QuiroHiro is our own clinic management software, designed to plug into HeyCAi natively. If you’re already running something else, the most common platforms in chiropractic and physiotherapy are PracticeHub, Cliniko, Jane App, Google Calendar, ChiroTouch and Genesis ChiroSoft. Check the assistant supports yours natively.
What to ask: native vs Zapier/Make (native is more stable), initial setup time, any extra cost per integration.
3. GDPR compliance and voice data protection
Any AI assistant handling phone calls in a European clinic processes personal health data. That triggers the strictest tier of GDPR (Article 9, sensitive data) and, in Spain, LOPD-GDD on top.
Minimum non-negotiables:
- Encryption in transit (TLS 1.2 or higher) and at rest (AES-256)
- Servers in the EU
- A retention policy with a defined maximum (90 days is reasonable for voice)
- A mechanism for patients to exercise access, rectification and deletion rights
- A signed Data Processing Agreement between your clinic and the vendor
What to ask: where data is hosted, how long recordings are kept, whether the vendor has a DPO, and whether the DPA is provided before you sign.
4. Real multilingual support, not machine translation
If your clinic gets international patients (tourists, expats, exchange students), you need an assistant that operates in more than one language. There’s a real gap between “multilingual support” and machine translation layered on top of a voice flow.
Genuine multilingual means: automatic language detection within the first seconds, clinical vocabulary adapted to each language (not literal translation), and the ability to switch language mid-call if the patient needs it.
HeyCAi currently handles Spanish and English natively. Other languages are on the roadmap, and clinics that need a specific language sooner can tell us and we’ll prioritise.
What to ask: how many languages are natively supported (not via real-time translation, which adds latency), whether clinical vocabulary is adapted per language, and whether you can customise greetings and sign-offs per language.
5. Customisation and clinical context
A generic assistant doesn’t know that in chiropractic a first visit takes 45 minutes and a follow-up adjustment takes 15. It doesn’t know that physio has group sessions and individual sessions with different durations. Or that in osteopathy many patients call with acute pain and need an urgent slot, not a routine check-up three weeks out.
The assistant should let you configure:
- Visit types with different durations
- Priority rules (emergencies vs routine)
- Schedules per practitioner (not everyone works the same days)
- Messages adapted to your clinic’s tone
What to ask: how much the conversational flow can be customised, whether there’s a dashboard where you can adjust schedules and visit types without depending on the vendor, and whether the assistant learns from your clinic’s specific context over time.
6. Pricing and billing model
Pricing models vary widely. The three you’ll see:
- Flat monthly fee
- Pay-per-minute of conversation
- Hybrid (base fee with included credits plus per-minute overage)
The hybrid model tends to be the most predictable for small and mid-sized clinics. Fixed fee with a reasonable volume of minutes included, overage only if you exceed it. Lets you budget without surprises.
Always compare total monthly cost against the real alternative. A part-time receptionist costs €12,000 to €16,000 a year all-in (with social security) and doesn’t cover nights or holidays. An AI assistant covering 24/7 for under €250/month has a clear ROI.
HeyCAi runs at €49/month for the WhatsApp-only plan and €229/month for the full Clinic plan (WhatsApp + voice 24/7).
What to ask: lock-in periods, cancellation terms, whether unused minutes roll over, and any hidden costs for integrations, support or updates.
7. Technical support and onboarding time
Onboarding is where most clinics get stuck. If configuring the assistant takes weeks of back-and-forth, the team loses patience and the project stalls.
Ask how long onboarding takes on average (not the best case, the real average). For a clinic with one management software and a standard booking flow, it shouldn’t exceed 5 business days. If they say “it depends”, ask for concrete examples of clinics similar to yours.
Post-implementation support matters as much as onboarding. A voice assistant is live software: it needs adjustments when you change hours, add practitioners or introduce new appointment types. Check whether support is by email, live chat or phone, and what the average response time is.
What to ask: how many clinics similar to yours are already live, what support channels they offer, whether there’s a dedicated account manager, and whether initial training is included.
Quick checklist: 7 questions to take to any vendor call
Before signing with any AI assistant vendor for your clinic, take this list with you:
- What’s the average latency in production (not in demos)?
- Is the integration with my management software native or via Zapier?
- Where is data hosted and how long are voice recordings retained?
- How many languages does it support natively?
- Can I configure visit types and schedules myself?
- What’s the total monthly cost, including any add-ons?
- How many days does onboarding take for a clinic like mine?
If the vendor can’t answer all seven clearly, that’s a red flag.
Frequently asked questions
What does an AI assistant for a clinic cost on average?
Prices range from €100 to €400 per month depending on the vendor, call volume and included integrations. Most common model: base fee plus included minutes. HeyCAi Personal starts at €49 for the WhatsApp side and HeyCAi Clinic at €229 with the voice receptionist included.
Is it legal to record patient calls with an AI assistant?
Yes, if GDPR is respected: inform the patient at the start of the call, get consent, define a legitimate purpose and limit retention. In Spain the LOPD-GDD adds extra requirements for health data.
How long does it take to get one running in my clinic?
Depends on integration complexity, but for a clinic with standard management software, onboarding is typically 3 to 7 business days. That covers calendar setup, appointment type configuration and live call testing.
What happens if the assistant doesn’t understand the patient?
Current voice assistants stay above 95% accuracy in both Spanish and English, including regional accents. If the assistant can’t resolve a query, it should escalate to a human automatically. Ask the vendor for the escalation protocol and what share of calls trigger it.
Can I try one before committing?
Many vendors offer demos or trial periods. With HeyCAi, you can trigger a real demo call to your phone in 30 seconds before deciding anything.
If you want to test these 7 criteria against HeyCAi specifically, book a demo call. For deeper questions about the voice side, HeyCAi Voice on callcai.ai goes into more detail.