Insights
By Ege Engin Özdaş (Co-Founder & CEO), Şener Özer (Co-Founder & CTO) & Gökçe Özyurt (Co-Founder & CPO) — Staterics · Published 2026-08-03 · Updated 2026-08-03 · 12 min read
The short answer: Operational AI for clinics answers calls and messages, books and reschedules, reports on device orders and repairs, takes accessory reorders, and runs post-visit follow-up. It works inside a scope the clinic writes down, and routes clinical judgement and anything sensitive to a person.
AI for clinics means two products that share a name. Clinical AI reads images, flags findings, supports diagnosis. That one belongs to your clinicians and their regulators. Operational AI answers the phone at 9:05 when four voicemails are waiting, books the fitting, says whether the device is back from repair, and posts the wax guards. This page is about the second one.
It is written for a business that is really two at one desk: a clinic that delivers care, and a retailer that sells devices, fits them, services them, and reorders the parts that keep them working. Most write-ups of this cover the booking half and stop. Orders, fittings, repairs, and reorders fill the same phone line and get almost none of the coverage.
Operational AI produces appointments, messages, and tasks, and is judged on what it is allowed to say and who can read the transcript. Within that, it splits again.
A point tool covers one channel. An AI operating system covers the desk: the system answering the call also sees the calendar, the device order, and the repair queue. For a clinic that also sells devices, that decides whether half its calls can be answered at all.
Clinical AI is judged on what it concludes. Operational AI is judged on what it is allowed to say.
An AI medical receptionist is software that answers a clinic’s calls and messages and completes the routine tasks they produce: booking, rescheduling, cancelling, taking a message, updating a record. It works from the clinic’s live schedule and rules, and acts only inside the scope the clinic approved.
One call, mechanically: it answers on the first ring, says it is automated, matches the caller to a record, reads real availability, writes the booking, sends the confirmation. Patients reach it by phone, message, or web chat, in the languages agreed during your build. More detail: what an AI receptionist actually handles.
A clinic day: four voicemails at 9:05, two of them cancellations for today; a patient at the counter while both lines ring; “is my hearing aid back from the lab?” The work is not difficult; it is simultaneous. Three piles.
Booking, rescheduling, cancelling. Reminders, and waitlist backfill when a slot opens. Device order status, the “is it ready yet” call. Repeat accessory reorders where item, size, and address are on file. Out-of-hours calls in the same categories.
Repair and warranty intake, with fault, serial number, and purchase date captured. Quotes and insurance questions, figures pulled but nothing promised. Clinical questions summarised for the clinician.
Clinical judgement of any kind. Symptoms that could be urgent. Distress, bad news, bereavement: anything where the caller needs a person, not an efficient one. Price negotiation, consent, commitments the business must honour.
A configuration, not a preference: which questions the AI answers, which it routes, what it says while routing. Set it during the build, with the refusals explicit. It never guesses at prices, medical advice, or commitments.
What stays human is not what is left over. It is what the system exists to protect.
Behind the clinical operation of a hearing centre, an optical shop, or an orthodontic practice runs a retail one: a device is ordered, fitted, adjusted, repaired under warranty, and consumes small parts forever. Every stage generates calls, and almost none are bookings.
Booking-first software cannot do this, and the reason is structural: “is my device ready” needs the order record; “can I collect two packs of domes” needs stock. Put those two calls to any vendor you shortlist.
Inbound answering is where this category starts and usually where it stops. What decides whether next quarter is full is outbound and scheduled: the six-month recall, the post-fitting check, the patient who bought a device in 2024 and has not been seen since. A person can build that list on a slow Tuesday, if nothing urgent happens first.
A system with one connected record has it already, sorted by date. The business stops forgetting. The limits are real: outbound needs consent, an agreed channel, and a do-not-contact list honoured without exception.
If you already have an online booking page, the fair question is what the newer thing adds.
| Online booking page | AI at the front desk |
|---|---|
| Serves the patient already on your site, at a screen, willing to complete a form. | Serves the channel the patient chose, including the caller who will never use a form. |
| A cancellation leaves a hole until a human notices. | A cancellation starts the waitlist while the slot is still worth filling. |
| Ends at the booking. Orders, repairs, and reorders live somewhere else. | One record covers the booking, the reminder, the device order, the repair, the reorder. |
The deciding factor is not the booking itself, which a good page handles well. It is everything after it: the booking page stops when the slot is taken, while one connected record carries the device order and the repair that the same patient will call about next.
Compliant is not a property software owns by itself. In the US the obligation sits with your practice as the covered entity, and extends to vendors handling protected health information, who sign a business associate agreement. Five things belong in writing:
In the EU, the UK, or Türkiye, HIPAA is the wrong statute, and a page mentioning only HIPAA has not answered you. GDPR, and KVKK in Türkiye, treat health data as a special category: expect a lawful basis, a data processing agreement, a retention period, and an erasure route.
You will not find performance percentages here. We do not publish client results as statistics. The numbers circulating in this category are rarely auditable, and never yours. Four failure modes are more useful, each with the thing that actually mitigates it.
Automation stalls exactly where the confusion already is. That is the most useful thing the first weeks will tell you.
AI employees are live within 14 days of data handover, and the complete operating system within 90 days, with no software migrations. The clock starts when we have your data, not when you sign. Past day 90 on the annual lane, the guarantee has an objective trigger: 100% of the deposit back, or pay the balance only on delivery. How long implementation takes covers the clinic’s side of the schedule.
What you commit in return is small and specific: the data handover itself, one named person who can settle policy questions inside a working day, and sign-off on scope before go-live. The build waits on those three, which is why the 14 days start where they do.
On the compliance list above, our own answers are fixed: trained only on your clinic’s data, kept private and never pooled, handed back if you leave, and acting only inside the authority levels you approved. We are not your compliance counsel. Recording and retention settings stay yours to sign off.
Four things move the price at any vendor: call volume, the share of calls completed end to end rather than merely assisted, how many locations run off one system, and how many systems the AI must read from. A calendar alone is cheap; a calendar plus an order and stock record is not. The full breakdown is in how much AI implementation costs.
Ours is published: a platform fee from $750 per month, set by system complexity and number of users, plus AI usage metered in RIC Tokens. One meter across voice and text, billed monthly in arrears on what the AI actually did. No development cost, and billing starts at go-live.
Voice runs about a token a minute, a chat exchange about 0.25, and the meter resets each month. The rate per token falls as monthly volume rises:
Two clinics, at volumes you can check against your own call log: a message-only clinic assistant at about 80 tokens a month, roughly $20 in usage; round-the-clock voice cover at about 3,000 minutes a month, $655 in usage, an effective 22 cents a minute.
Count for yourself before comparing any of that to a salary. From last week: how many inbound calls were a booking, a reschedule, an hours question, an order-status check, or a reorder? That is the automatable pile. How many needed a clinician? That pile never moves. The comparison with hiring another front-desk person follows.
Or have the map drawn for you: the free 45-minute Operations X-Ray walks through the callback list, the workshop tray, and the reorder texts, and leaves you with a friction map and the three workflows to take off the desk first.
Software that answers the clinic’s calls and messages and finishes the routine tasks they create: booking, rescheduling, hours questions, device order checks. It reads the live calendar and the clinic’s own records, and routes anything outside its approved scope to a person.
It should be built around the tools you already run on, with no migration. Name yours first: hearing care runs on Noah, Sycle, or Blueprint; general clinic practice on the likes of Cliniko or Dentally. Then sort them into two tiers. Some publish developer documentation for a direct integration; others are reachable only through a partner interface or a local desktop module, which needs a shared-record handoff, where the AI writes the entry and a person carries it across. Check your platform’s current developer documentation, and ask which tier each of your systems is in before you sign.
Compliance is a property of the arrangement, not of the software. In the US, ask whether the vendor signs a business associate agreement, which sub-processors it covers, whether calls are recorded or only transcribed, where transcripts live and for how long, and who can read them. In the EU, the UK, or Türkiye, HIPAA does not apply: ask for a data processing agreement, a lawful basis under GDPR or KVKK, and an erasure route.
An AI receptionist never gives medical advice or assesses symptoms. The configuration defines what triggers immediate escalation: in hours, a live transfer to a named person; out of hours, the clinic’s own emergency script and a flagged record waiting when it opens.
Yes, provided the AI sees the order record and stock and not only the calendar. That is the deciding capability for a clinic that sells devices. It covers order-status calls, fitting and trial-period check-ins, repair intake, and reorders where item, size, and address are on file.
Some patients will not use an automated voice, and that is a design input rather than an objection to argue with. Say in the first sentence that the assistant is automated, offer one obvious route to a person, and transfer after a second failed exchange rather than retrying.
In Staterics builds, AI employees are live within 14 days of data handover and the complete operating system within 90 days, with no migrations. Billing has two lines: a platform fee from $750 per month depending on complexity and number of users, plus usage metered in RIC Tokens, $0.25 down to $0.16 per token as volume rises.
Clinical judgement, conversations carrying distress or bad news, price negotiation, consent, and any commitment the business must honour. The AI never guesses at prices, medical advice, or commitments; it routes those calls with everything collected.
Yes: one patient record across sites, one calendar per clinician, one usage meter. Each location keeps its own hours, staff, stock, and phone number, and the routing rule you set decides whether a caller is offered the nearest site or the earliest slot. Adding a site is a configuration change, not a second deployment.
The same categories handled end to end in hours are handled out of hours: booking, rescheduling, hours and address questions, order-status checks, reorders. A clinical emergency follows the clinic’s own script, never medical advice, and anything outside approved scope leaves a flagged record waiting when the clinic opens. If you already pay an answering service, compare the two per category: which calls each one finishes, and which it only takes a message for.