--- title: "An AI Agent in Healthcare: Scheduling and Administration" description: "In healthcare the line is clearer than anywhere else: an agent may handle administration, not anyone's health. Where it genuinely helps and what has to be settled first." canonical: "https://interfase.sk/en/news/an-ai-agent-in-healthcare-scheduling-and-administration" lang: "en" last_updated: "2026-10-26" --- # An AI Agent in Healthcare: Scheduling and Administration > In healthcare the line is clearer than anywhere else: an agent may handle administration, not anyone's health. Where it genuinely helps and what has to be settled first. Published: 2026-10-26 - An agent in healthcare must not assess symptoms or recommend a course of action — it may handle scheduling, reminders and administration. - Health data is a special category of personal data and demands a stricter regime than ordinary business automation. - The largest saving is on the phone: most calls to a clinic are bookings, changes and repeated questions. - Any enquiry that sounds like an acute condition has to go to a person immediately, with clear instructions on where to call. The phone in a clinic rings all morning. A nurse answers between patients, some calls go unanswered, and those people try again an hour later. Most of them are calling to book, to move an appointment, or to ask about opening hours. This is exactly the kind of load that can be lifted. At the same time, healthcare is an environment where the limits need setting more strictly than anywhere else. ## The line: administration yes, medicine no Start with what the agent **must not** do. It must not assess symptoms, estimate urgency from a description of complaints, recommend a course of action, or do anything resembling health advice. That is not only a matter of regulation — it is a matter of patient safety. What it **can** do is the whole administrative layer around it: booking, rescheduling, reminders, information on how to prepare for an examination, opening hours, what to bring. The line fits in one sentence: **the agent may talk about organising the visit, not about anyone's health.** ## An acute condition has only one right answer Even with a purely administrative agent, an enquiry will arrive that sounds serious. A fixed response has to be ready for that, not improvisation by the model. The rule is simple: at any mention of a condition that may be acute, the agent **stops handling the booking**, points clearly to the emergency number or urgent care, and passes the contact to staff. It does not ask for details, does not reassure, does not assess. This response is configured as a hard rule outside the model — not as an instruction in a prompt that can be forgotten. ## Health data carries a stricter regime Under GDPR, data about health is a special category. In practice that means procedures which suffice for ordinary business automation do not suffice here. What has to be settled: **where the data physically sits** and whether it leaves the space it should stay in; **who can reach it** and with what record; **how long it is kept**; and **what should not be recorded at all** — the content of health-related questions does not need to enter a call transcript. We covered the basic principles in the piece on [AI agents and GDPR](/blog/ai-agenti-a-gdpr-na-co-si-dat-pozor-pri-spracovani-osobnych-udajov); in healthcare, professional confidentiality and the provider's own internal rules come on top. ## Where the saving is largest **Booking and rescheduling.** Finding a free slot, moving an existing appointment, recording it in the system. This is the bulk of calls and the agent handles it without intervention. **Reminders before appointments.** Reducing no-shows is a measurable and quick effect. A reminder the day before, with one-click cancellation, frees the slot for someone else. **Repeated questions.** Opening hours, where to park, whether to come fasting, which documents to bring. The answers come from the clinic's own documents, not from the model's free reasoning. **Routing enquiries.** Telling apart what belongs to a nurse, what to a doctor and what is administrative — without assessing the content of a medical question. ## Not every patient wants to type The age profile of patients in healthcare differs from an e-shop's. A text chat cannot be the only route — the phone has to stay, and reaching a person must not be obstructed. A good deployment reduces call volume because some people book for themselves, which means everyone else gets through faster. A bad one puts an obstacle between the patient and the nurse. ## What to measure - **Share of bookings completed without staff.** - **Number of missed calls** — before and after. - **No-show rate** for appointments. - **Staff time spent on the phone.** And one check that matters more than the rest: **how many enquiries were correctly handed to a person**. If that number falls, the agent is taking on more than it should. ## Where to start With appointment reminders. There is no judgement involved, the benefit is visible within a month, and staff get used to the tool before booking comes into play. If you are considering something similar and want to go through what is and is not permissible in this environment, [get in touch](/kontakt). More on our approach to [AI and automation](/riesenia/ai) and examples in our [case studies](/referencie). --- Web version: https://interfase.sk/en/news/an-ai-agent-in-healthcare-scheduling-and-administration llms.txt: https://interfase.sk/llms.txt