A hospital chatbot answers appointment, report, bill and doctor-availability questions on WhatsApp. It does not do triage and shouldn't try.
What WhatsApp actually solves
WhatsApp is, by a wide margin, the most-used app in Indian healthcare communication — between patients and front-desk staff, between doctors and patients, between branches of the same hospital. A WhatsApp-based chatbot replaces the 'send message to receptionist's personal number' workflow with a single hospital-owned WhatsApp business number that handles structured queries 24x7 and hands off to a human when the bot doesn't know. The unlock is consistency — every patient gets the same response, the response is logged, and the receptionist's personal phone stops being a hospital asset.
The five conversations every hospital chatbot should handle
- Book an appointment — specialty, doctor, preferred date, OTP-confirmed.
- Reschedule or cancel an existing appointment — OTP-confirmed.
- Lab report status — 'is my report ready?' with a one-click link to download the PDF.
- Outstanding bill enquiry — current balance, last payment, payment link.
- Doctor availability — 'is Dr X available today?' with the next slot offered for booking.
What it should never try
Symptom triage. Medical advice. Diagnosis. Drug recommendations. Lab interpretation. Anything where a wrong answer could harm a patient. A chatbot that says 'your symptoms sound like flu, take paracetamol' is a malpractice case waiting to happen, and no scripted guardrail is robust enough to prevent it once you put a freeform language model in front of it. Keep the chatbot's scope deliberately narrow — administrative, not clinical.
Handover to a human
Every chatbot should have an explicit 'I don't know, let me put you in touch with the front desk' path that triggers within 2 minutes of an unrecognised query, hands the conversation transcript to a human receptionist, and gives the patient a clear ETA. The handover quality is what separates a chatbot that patients trust from one they bypass. Design for it from day one.
Multilingual but bounded
Patients write in Hinglish, Marathi, Tamil — sometimes within the same conversation. The chatbot should detect the language, respond in the same one, and not insist on English for structured fields. But the response templates should still be authored by humans, in each language, and reviewed for tone. Auto-translated responses sound like they came from a 2010 Google Translate, and patients trust them less.
What we measured
- Front-desk message volume on personal numbers dropped by ~70% after WhatsApp chatbot adoption.
- Patient satisfaction with response speed rose — chatbot median response is 6 seconds vs receptionist median of 14 minutes.
- Off-hours bookings (between 8 PM and 8 AM) added 8-11% to monthly OPD volume — these slots simply didn't exist before.
- False-positive escalations (chatbot escalated when it shouldn't have) ran at ~9% in month 1, dropped to ~3% by month 3 as the template library matured.