Hospitals · AI agents and chatbots
AI agents and chatbots for hospitals.
Reviewed 2026-09-25Read as markdown
The market
How hospitals in India actually operate.
Most Indian hospitals below the corporate chains still run on a patchwork: a billing package bought years ago, registers for the wards, spreadsheets on the insurance desk and WhatsApp for everything in between. The expensive failures happen in the gaps — a pre-authorisation stalled for one missing document, a discharge held up by a bill nobody can finalise, a patient registered twice because the record lives in another department.
Workflow
The sequence we build around.
Registration and the hospital number
Every visit should hang off one number for the patient. In practice someone seen in OPD, sent for a scan and admitted a week later often ends up with three records, because each department registered them in whichever system it happened to have open.
OPD consultation and orders
The consultant’s orders — investigations, medicines, a referral — are where billing starts. When they are written on paper and re-keyed at the counter, the charge that goes missing is usually the one the hospital was least likely to notice.
Admission, beds and the running bill
Once a patient is admitted the bill grows by the hour: bed category, procedures, consumables, pharmacy issues. Families ask for the running total every day, and a system that can only produce it at discharge turns each of those questions into an argument.
Insurance pre-authorisation and the TPA
For an insured patient the cashless request, the query that comes back, the enhancement mid-stay and the final approval at discharge each need documents from three departments. The insurance desk is where a hospital’s paperwork problem becomes a revenue problem.
Discharge and follow-up
The discharge summary, the final bill and the follow-up appointment all have to finish before the bed is free. The summary is usually the bottleneck, because it waits for a doctor to write it from notes scattered across the whole stay.
Integrations
What it has to talk to.
- The OPD schedule and appointment system
- An agent that can only take a phone number and promise a callback has moved the queue, not cleared it. Reading consultant availability and writing the booking into the schedule the OPD runs from is what turns an enquiry at eleven at night into an appointment.
- The tariff, package and insurer lists
- Most enquiries are about cost and cover: what a package includes, whether a policy is accepted cashless, what deposit admission needs. Grounding the agent in the hospital’s own maintained lists, not in general knowledge, stops it quoting a figure the billing counter will not honour.
- The patient record, read-only and scoped by role
- The insurance-desk agent needs the diagnosis, the estimate and the investigations to assemble a pre-authorisation file; the patient-facing agent needs none of it. Access is scoped per agent, logged, and never wider than the task in front of it.
- The hospital’s WhatsApp Business number
- Families of admitted patients ask the same things every day — visiting hours, the running bill, what to bring at discharge. Answering on the channel they already use, with a handover to the ward desk for anything about the patient, takes those calls off the nursing station.
Constraints
What changes the build.
- Rates have to be displayed
- Clinical establishment rules in many states require a hospital to display what it charges for each service, in English and the local language. A billing system whose tariff lives in one place can publish the same figures to the counter, the estimate and the website instead of three versions that drift apart.
- Insurance decisions now run on a clock
- The insurance regulator’s 2024 master circular on health insurance puts clocks on insurers: a decision on a cashless request within an hour, and final approval within three hours of the hospital’s discharge request. A file sent incomplete comes back as a query, and the patient waits in the bed while the gap is filled.
- ABDM is optional until it is not
- Registering the facility on the Health Facility Registry and linking records to a patient’s ABHA number is voluntary for most private hospitals, but it is increasingly expected by government schemes and payers. A record designed with ABHA fields and consent flows from the start avoids a rebuild later.
- Some registers are statutory
- An obstetric ultrasound needs a Form F under the PC-PNDT Act, controlled drugs need their own register, and a medico-legal case needs its records kept for as long as the case can run. These are inspected, so they cannot be an afterthought bolted onto a billing package.
- The hospital cannot stop when the internet does
- Registration, the emergency desk and the pharmacy run through the night and through outages. A system that only works while the cloud is reachable is one the night staff will quietly replace with a register, so the counters that cannot stop need a local fallback that catches up when the connection returns.
Scope and timeline
What a build involves, and how long it takes.
| Scope | Typical timeline |
|---|---|
| Patient enquiry and appointment agent grounded in departments, packages and insurers | 4 – 7 weeks |
| Agents wired into the patient system: insurance files, discharge follow-up and family updates | 8 – 14 weeks |
For a quote, tell us what you need. We come back with a fixed scope and a fixed price once we have understood it.
A worked example
One engagement, in three parts.
- Context
- The nearest agent we have shipped in healthcare is the assistant on the Curis 360 site, built for a three-clinic physiotherapy practice that was losing enquiries sent after the front desk had gone home.
- What was built
- An assistant that answers only from the practice’s own pages — what a condition involves, how long a course usually runs, what it costs, which clinic is closest — takes the booking at the end of the conversation, and sends anything clinical to a physiotherapist with the thread attached.
- What changed
- Messages that arrived at night began ending in appointments instead of a morning callback list, and clinical questions reached a clinician rather than being answered by software. A hospital runs the same design per department, and the handover rule carries more weight there, not less.
What goes wrong
The mistakes we see in hospitals.
- Letting the agent answer from general medical knowledge. Someone asking whether their symptoms need the emergency department needs a clear instruction to call or come in, not a paragraph of possibilities; the agent should recognise the question and route it at once.
- Giving one agent access to everything. The agent that books appointments has no reason to read clinical notes, and a hospital that cannot say which agent opened which record cannot answer a data-protection request honestly.
- Quoting package prices from a document last updated a year ago. The agent is only as current as the list it reads, so it has to read the same maintained tariff the billing counter uses.
Questions
What hospitals ask.
Can an AI agent handle emergency enquiries for a hospital?
It should recognise them, not handle them. Any message that reads like an emergency gets the emergency number, the address and an instruction to come in — immediately, without a conversation. That rule is written into the agent before anything else about it is designed.
Can the agent give families updates about an admitted patient?
Only what the hospital decides can be shared, and only after the family member is verified. General questions — visiting hours, parking, what to bring at discharge — it answers directly. Anything about a patient’s condition goes to the ward desk or the treating team, with the question attached.
What does an agent do at the insurance desk?
It assembles. For a cashless request it pulls the diagnosis, the estimate and the investigation reports from the record into the insurer’s format, lists what is missing, and reminds the right department to supply it. A person reviews and submits; the agent removes the chasing, not the judgement.
Does a hospital agent need to work in regional languages?
Usually, yes. Patients and families often write in Kannada, Hindi, Tamil or Telugu, or in a mix of English and their own language. The agent replies in the language of the message, while the content it answers from stays one maintained source so the facts do not drift apart between languages.
Will an AI agent replace the hospital’s front desk?
No. It takes the repetitive questions and the overnight queue so that the desk can deal with the people standing in front of it. Every conversation it cannot resolve arrives at the desk with the question, the answer given so far and the patient’s details already attached.
This page is also available as clean markdown with no navigation or styling, for anything that would rather read it that way. Last reviewed 2026-09-25 by Neel.
Where to next
- AI agents and chatbots, in general
How MyFloww approaches AI agent and chatbot development for any business, not only hospitals.
- Everything we build for hospitals
The industry hub: the workflow, the constraints and the other services we publish for it.
- AI agents and chatbots for physiotherapy clinics
The same service in the nearest industry, where the constraints differ.
- Apps and custom software for hospitals
Customer apps for Android, iOS and web, and the internal web application a business runs on — bookings, payments, inventory, staff, custom CRM and ERP.
- Next.js websites for hospitals
Hand-written Next.js sites with no theme or page builder, built to rank in Google and be quoted by AI answer engines.
Building AI agent and chatbot development for a hospital?
Tell us what you are trying to do and we will come back with a scope and a number. AI chatbots trained on your own content, and agents that carry out the multi-step work behind the conversation, with an explicit hand-over point to a person.
Or message us on WhatsApp (opens in a new tab) · info@myfloww.in