Guide
How to choose an AI receptionist for your hospital: the questions to ask before you buy
There is no single best AI receptionist for Indian hospitals. These are the questions to ask any vendor, what a good answer sounds like, and the red flags.
In short
There is no single best AI receptionist for every Indian hospital: the right one is the one that proves itself on calls like yours. Before you buy, ask any vendor for results from a live hospital line, and test for yourself how it handles an emergency, a medicine question, a fee it was never given and a caller who wants a person, in your patients' Hindi or Hinglish, with bookings landing in your own diary and patients calling your own number. Then get the data terms in a contract under India's DPDP Act, and a written quote for a full year at your own call volume.
- Ask for results from a live hospital line, with dates and how each number was counted. A demo proves little.
- Test emergencies, medicine questions, unknown fees and 'are you a person?' yourself, on your own setup.
- Patients should keep calling your number, and bookings should land in the diary your counter uses.
- From 13 May 2027 the DPDP Act holds your hospital responsible for what the vendor does with patient data, so put it in a contract now.
- Compare prices as a written quote for a full year at your own call volume.
A list that ranks AI receptionists can’t tell you how one will cope with your hospital’s calls: your doctors, your callers’ Hindi, your morning rush, your nights and Sundays. Questions can. This guide gives you the questions to put to any vendor, why each one matters, what a good answer sounds like, and what should make you walk away.
It names no products. We make one, DrMate, so our own answers are at the end, held to the same standard.
Which questions should you ask an AI receptionist vendor?
Ask every vendor the same twelve questions and keep the answers side by side; the sections below explain why each matters.
| Ask | A good answer | A red flag |
|---|---|---|
| 1. Can you show results from a live hospital line? | Counts and dates from real calls, and how each was counted | Only a demo, or a percentage with no dates and no call count |
| 2. What happens when a caller mentions chest pain halfway through a booking? | Every sentence is checked; the caller is told what to do now (108, 112 or your emergency) and staff are alerted | “The AI will understand”, or a transfer to reception with no instruction to the caller |
| 3. How is it stopped from giving medical advice or inventing a fee? | It answers only from facts you approved, and the system checks before anything is said | “Our model is trained not to” |
| 4. Will it say it is an AI? | Yes, whenever a caller asks; it never claims to be a person | “Callers can’t tell the difference” |
| 5. Can we test it with our own doctors, in our patients’ Hindi? | A test number with your doctors and fees loaded, and every test call’s transcript | A polished demo, or a long list of languages |
| 6. Will patients keep calling our number? | Yes: your line forwards to it, starting with missed and after-hours calls | A new number for patients to learn |
| 7. Where does a phone booking land? | In the diary your counter and doctors use, confirmed only once it is saved | A request your staff must confirm later |
| 8. How does a caller reach a person? | Staff phones you choose ring in hours you set; if nobody answers, a call-back list with the reason | No hand-over, or calls that end with nothing written down |
| 9. Who fixes a wrong answer, and how is the fix checked? | Every call has a transcript; each mistake becomes a test it must keep passing; updates avoid OPD rush hours | “It learns by itself” |
| 10. Who is responsible for patient data? | You are; the vendor processes it for you under a data processing agreement | “Our terms of use cover it” |
| 11. Are calls recorded, how long is call text kept, and where is it hosted? | Plain answers: recording yes or no, days kept, where hosted, who can see it | Vague answers: “in the cloud”, “as long as needed” |
| 12. What will a year cost at our call volume? | A written quote for a year, with everything extra listed | A per-minute rate and nothing else |
Why ask for results from a live line, not a demo?
A demo runs on a good microphone, with a made-up patient who speaks clearly and never changes her mind. Your line has a son calling from a busy market for his father, a grandmother who pauses between words, and calls before the counter opens. The best proof a system copes is what happened on a real hospital line. Ask for:
- Calls answered over a stated period, test calls left out.
- Calls outside OPD hours, the ones a day-time desk can’t take.
- Calls that ended in a booking the system made itself, not a request for your staff to confirm.
- Calls passed to a person, and why. None at all should worry you as much as too many.
- Callers who hung up before the end, and whether anyone rang them back.
- How long calls take, and how quickly it replies.
Here are ours: DrMate’s first 17 days on one hospital’s OPD line, 29 October to 14 November 2026 (full results).
| What we counted | Result |
|---|---|
| Calls answered by DrMate | 76 |
| Before 9 am or after 8 pm | 21 of 76 (28%) |
| On Sundays | 6 of 76 |
| Ended with an appointment booked | 28 of 76 (37%); DrMate booked 32 appointments in all |
| Passed to a person at the hospital, because the caller asked for one or it was urgent | 7 of 76 (9%) |
| Hung up before finishing | 38 of 76, 19 of them within 15 seconds; 31 of the 38 went on the front desk’s call-back list |
| Length of a call | About a minute for a typical call (median 61 seconds); about two minutes for one that ended in a booking (median 2 min 6 s) |
| Time from the caller’s last word to the reply | About a second (median 0.9 seconds) |
The numbers are small: 17 days, and only the calls passed on from one of the hospital’s OPD lines. They show what reached DrMate, not every call the hospital got. Expect the same from any vendor: dates, counts, the base of each percentage, and the limits.
How is safety enforced on every call?
An emergency doesn’t always come first. A caller booking a routine review may mention chest pain halfway through, almost in passing. So ask whether warning signs are checked on every sentence the caller says, or only at the start.
Then ask what happens at 2 am, when nobody is at your desk. An alert to reception helps only if someone sees it, so the instruction to the caller has to work on its own: call 108 or 112, or come to the emergency now. Talk of self-harm needs a kind reply and Tele-MANAS, the government’s free mental health helpline on 14416, open round the clock. Ask, too, who signs off what it says in an emergency: your medical superintendent or director, not the vendor.
The same goes for what it must never do: diagnose, suggest a medicine or dose, invent a fee or a doctor’s hours, or claim to be a person (what an AI receptionist must never do). Ask how each is enforced, then call and try to make it slip.
Will your callers be understood?
Callers don’t speak textbook Hindi. They mix in appointment, report and fee; they say “परसों” (the day after tomorrow, or the day before yesterday) and “lunch के बाद” (after lunch); they read their number out in pieces.
A long language list often shows only what the speech engine can turn into text, not whether the receptionist can book, read back and raise the alarm in each language. Ask in which languages the whole booking and the emergency checks were tested on real phone calls. Then test it yourself over an ordinary phone line, with your own doctors and fees loaded, and have your staff and a few older people call too. Ask for the transcript of every test call. Our 15-call test in Hindi takes one afternoon.
Does it fit your number, your diary and your staff?
Your number. Patients should keep calling the number on your board and prescriptions, forwarded to the AI, starting with the calls your team misses and those after hours. A new number for patients is a mistake twice over. It confuses them, and if you ever change vendors, the number on your posters may not be yours to keep. With forwarding, you simply switch it off.
Your diary. A phone booking should land in the same diary your counter and doctors use. If the system only takes requests for your staff to confirm, every booking is handled twice.
Your staff. Every caller it can’t help should reach a person or the call-back list, with the reason. A question your team answers once should be used from then on.
Bad days. Ask how many callers it takes at once, what the next one hears, and where calls go if the service is down; who fixes a call it gets wrong, how quickly, and how they check the fix hasn’t broken something else. A system that changes what it tells patients unchecked is a risk, not a feature.
Who is responsible for patient data under the DPDP Act?
This is general information for hospital teams, not legal advice.
Under the Digital Personal Data Protection Act, 2023, your hospital is the data fiduciary, which decides why and how patients’ data is used (section 2(i)). The vendor is a data processor, handling it on your behalf (section 2(k)). Two lines matter most when you buy: you stay responsible for following the Act in everything the vendor does with your data, “irrespective of any agreement to the contrary” (section 8(1)), and you may use a processor “only under a valid contract” (section 8(2)).
These duties apply from 13 May 2027, eighteen months after the government’s notification of 13 November 2025 (Gazette of India, G.S.R. 843(E)). From then, failing to keep reasonable security safeguards can cost up to ₹250 crore (PIB, 2025). A contract signed now will still run then, so ask:
- Will you sign a data processing agreement covering security, deletion and breaches? The DPDP Rules, 2025 require your contract with a processor to provide for security safeguards (rule 6(1)(f)). After a breach, your hospital must tell the Data Protection Board and every affected patient (section 8(6)): both without delay, and the Board must get a detailed report within 72 hours unless it allows longer (rule 7). So the contract should make the vendor tell you at once.
- Are calls recorded? If so, are callers told first, who can listen, and for how long?
- How long is the text of each call kept, and are any words kept longer, and why?
- Where is the data hosted, and what leaves India? The Act doesn’t require data to stay in India: it lets the government restrict transfers to countries it notifies (section 16), and other laws can be stricter.
- Who at the vendor can see your patients’ details, and is that access logged?
- What happens when you leave? The Act asks you to erase patients’ data when they withdraw consent or its purpose is no longer served, and to have your processor erase what you gave it, unless a law requires keeping it (section 8(7)). Put deletion when you leave, confirmed in writing, in the contract.
Get the answers in writing; a badge is not an answer. Our DPDP guide for hospitals covers the rest.
How do you compare prices that don’t look alike?
Quotes can come per minute, per doctor or as a flat monthly fee, sometimes with an allowance and a rate for anything over it. You can’t compare them until you turn each into one figure: what you would pay in a year at your own call volume.
As an illustration, take 100 calls a day lasting 90 seconds on average (example figures, not the line above). That is 150 minutes a day (100 × 90 ÷ 60), or 54,750 minutes a year if calls come every day (150 × 365). Take your own numbers from your phone records. Then ask of every quote:
- How is a minute counted? On the line above, the median call took 61 seconds, so at least half the calls ran past a minute. Billed by the started minute, each of those counts as at least two.
- What is included? Charges for the forwarded calls (ask your own telephone provider), set-up, updating doctors and fees, enough lines for your busiest morning, hand-over to staff, WhatsApp messages and support hours.
- What happens at the end? The notice period, whether the price can rise, and what happens to your data.
- What comes back? Bookings made on the phone, and calls answered after hours. Cost alone is half the story.
Where DrMate fits
DrMate is an AI front desk for Indian hospitals, built in India by TuringEdge. Our answers, in short:
| Question | DrMate’s answer |
|---|---|
| Live results? | 76 calls answered in its first 17 days on one hospital’s OPD line, with every figure on our results page |
| Emergencies? | Every caller sentence is checked for warning signs first; the caller is told what to do now (108, 112 or your emergency), staff see a red alert, and it never books a routine slot for an emergency. A safety net, not triage |
| Advice and facts? | It never diagnoses or suggests medicines, answers only from facts you approved, and says it is your hospital’s AI receptionist when asked |
| Language? | Everyday Hindi and Hinglish. It understands callers who speak English but answers in Hinglish, and doesn’t speak other Indian languages on the phone |
| Number, diary, staff? | Your line forwards to it; bookings land on DrMate’s front desk, used alongside your HIS; with hand-over on, it rings up to two staff numbers in the hours you set, else the caller goes on your call-back list |
| Mistakes? | Every call has a transcript, each mistake becomes a test it must keep passing, and updates avoid OPD rush hours |
| Data? | Built for India’s DPDP Act, 2023: you stay the data fiduciary; we process data only for you, under a data processing agreement. Calls are recorded only if you switch recording on, with callers told at the start; call text and recordings are deleted after 30 days, backup copies within 30 more, and to fix a mistake we may keep a call’s words with names and numbers removed. Hosted in India, on Microsoft Azure; our partners for phone calls, speech and WhatsApp handle only what their part needs and are named in the agreement (safety and privacy) |
| Price? | We price each hospital from its own call numbers and the modules it picks, and send a proposal that shows what you pay against what you get back |
| Going live? | 1 week; your team tests it on a test number before any patient calls |
Test it yourself on our demo line, 080 6553 5931, where it answers as Sunrise Hospital, a sample hospital we made up, or ask for a demo with your own doctors and timings.
Read this guide in Hindi (हिंदी में पढ़ें).
Questions people ask
Which is the best AI receptionist for hospitals in India?
There is no single best one for every hospital. The right one speaks the way your patients speak, handles emergencies safely, books into your own diary on your own number, and can show results from a live hospital line. Ask every vendor the same questions and test each one on your own setup before you decide.
What should we ask an AI voice agent company before buying?
Ask for results from a live hospital line; what it does when a caller describes an emergency; how it is stopped from giving medical advice or inventing a fee; whether you can test it with your own doctors; where its bookings land; how callers reach a person; who is responsible for patient data; and what a year will cost at your call volume.
How is an AI receptionist priced?
Quotes can come per minute, per doctor or as a flat monthly fee, sometimes with set-up, telephone charges and messages on top. Ask each vendor for a written quote for a full year at your own call volume, and ask how a minute is counted.
Should an AI receptionist take all our calls from the first day?
No. Let your staff test it on a test number first, then start with the calls your team misses and the ones after hours, on your own number. Hand it more calls once you have read a few weeks of its transcripts.
Does an AI receptionist need to record calls?
No. Your team needs the words of each call for a while, to check its work, but keeping the audio is a separate choice. If calls are recorded, callers should be told before recording starts, and you should know who can listen and when the recording is deleted.
Sources
- The Digital Personal Data Protection Act, 2023 (No. 22 of 2023), Gazette of India, 11 August 2023. Ministry of Law and Justice (copy hosted by the Telecom Regulatory Authority of India), 2023.
- Notification G.S.R. 843(E): dates on which the provisions of the DPDP Act, 2023 come into force (Gazette of India, 13 November 2025). Ministry of Electronics and Information Technology, Government of India, 2025.
- Digital Personal Data Protection Rules, 2025 (G.S.R. 846(E), Gazette of India, 13 November 2025). Ministry of Electronics and Information Technology, Government of India, 2025.
- DPDP Rules, 2025 Notified: A Citizen-Centric Framework for Privacy Protection and Responsible Data Use. Press Information Bureau, Government of India, 2025.
- Tele MANAS: over 14.7 lakh calls served in two years. Press Information Bureau (Ministry of Health and Family Welfare), Government of India, 2024.