Using AI in your clinic without losing the plot

Every week another product promises to "transform" your practice with AI. Most clinics don't need transformation. They need the phone answered, the appointment book kept straight, and the doctor freed to spend the consultation looking at the patient instead of the screen.

Start with the work nobody went to medical school for

The safest, highest-return uses of AI today sit at the edges of care, not the centre: answering routine calls and booking slots, sending reminders and follow-up messages, drafting notes for the doctor to review, and summarising a long history before a visit. If the AI gets one of these wrong, a human notices and fixes it before harm is done.

Keep the decision with the clinician

Diagnosis, prescribing and triage of serious symptoms carry real risk. Tools that touch these should be framed as suggestions, show their reasoning, and make it easy to disagree. A good rule:

If you would not sign it without reading it, the AI should not send it without you reading it.

Questions to ask any vendor

Where does patient data go? Ask which servers, which country, and who can access recordings and transcripts. What happens when it is unsure? A voice agent should hand over to staff, not guess. Can I see it fail? Ask for a live demo with awkward calls — a patient speaking Hindi and English mixed, a caller describing chest pain. What does it cost per month at my volume? Per-minute pricing adds up quickly in a busy OPD.

Measure, then expand

Pick one task, run it for a month, and count: missed calls, no-shows, time saved, patient complaints. Expand only what clearly helped. AI earns its place in a clinic the same way a new staff member does — by being reliable on small things first.

Disclosure: the author is the founder of Curovana, which builds clinic software including an AI voice receptionist. This article is general guidance, not an endorsement of any product.