When a healthcare institution considers an AI phone assistant, the first question is usually not what it can do, but where it stops. The institution’s phone carries sensitive conversations, so the limits have to be clear before the first call. In this article we describe them using AiMED Voice as the example.
One rule: admin, not decisions
AiMED Voice is an administrative tool. It answers the phone, books appointments and answers organizational questions. Medical decisions stay with the doctor. Everything else in this article follows from that one rule.
What the assistant may do
Answer every call. By day, at night, on weekends and on holidays, in Bosnian, Croatian, Serbian and 30+ more languages. Calls outside working hours are marked separately in the call log.
Book, reschedule or cancel an appointment. The assistant finds open slots by service and doctor, taking working hours and absences into account. It books an appointment only once the caller confirms, and the appointment appears in the portal calendar right away.
Answer organizational questions. Prices, services, address, working hours and preparation for an exam. The assistant answers from the knowledge base the institution enters in the portal, and it knows only what the institution enters.
Check who it is talking to. If the number is known, the assistant checks whether it is talking to that person. If not, it asks for a first name, last name and date of birth. It does not disclose patient data until the identity is confirmed.
Log the call. Every call is in the log with its time, caller, reason, outcome and duration, along with an AI summary, transcript and recording.
What the assistant must say
At the start of every call the assistant says that it is an AI assistant and that the conversation is recorded. Nothing is presented as a conversation with a person, so the caller always knows who they are talking to.
This is not a formality. A caller who knows they are talking to an assistant also knows what to expect from the conversation: an appointment and information, not a medical opinion.
What the assistant may not do
- It does not make diagnoses. It does not interpret symptoms and does not advise on therapy or doses.
- It does not assess urgency. When a caller says they have urgent symptoms, the assistant does not advise or assess. It tells them to call emergency services on 112.
- It does not read reports. It has no access to medical records: reports, medical histories or imaging.
- It does not pretend to be a human. It says it is an AI assistant, in every call.
- It does not answer beyond the knowledge base. When it has no answer, it says it doesn’t have that information and offers to have the institution call the caller back.
The last point matters in practice. The caller gets either an answer the institution wrote itself, or a clear message that the assistant doesn’t have that information, with an offer of a call back from the institution.
What stays with staff
The assistant takes over answering, booking and the questions that repeat. What stays with staff is what needs a person.
- Medical questions. Everything about symptoms, therapy and reports stays with the doctor.
- Call backs. When the assistant didn’t know something, you see the reason for the call and a summary in the call log, so you know whom you are calling and why.
- The knowledge base. The institution decides what the assistant knows. Once you add to the knowledge base, the assistant knows the answer from the next call.
- The calendar. The institution enters the services, doctors, working hours and absences. The assistant books only into open slots of that calendar.
- Access. In the portal each person has their own access, and the institution’s administrator decides who can change what.
The limits apply to data too
Only the institution the patient called can see the recordings, transcripts and summaries, and every view of a call is recorded in the access log. The conversation recording is deleted 30 days after the call, and the call content after six months. At the patient’s request the data is deleted permanently.
Speech and AI are processed in the EU, and the data processing agreement is signed before the first real data. More on this on the Security page.
Questions worth asking before you start
These are questions you can put to any provider of a phone assistant, including us.
- What does the assistant say at the start of a call?
- What does it say when a caller talks about symptoms or asks about therapy?
- What happens when the answer is not in the knowledge base?
- Who can see call recordings, and how long are they kept?
- Where is the data processed?
- How much of all that can staff see after the call?
The answers for AiMED Voice are in this article. If you don’t get a clear answer to one of them, that is an answer too.
To see one call from pickup to the record in the portal, watch the animated presentation on the How it works page. If you would like to see the assistant with your services and calendar, request a demo.
