AI receptionist vs IVR vs call centre: which suits a hospital OPD

A hospital that wants to fix its phones usually ends up choosing between three things: a better IVR, more people on the phones (its own or an outsourced team), or an AI receptionist. Each is the right answer for some calls and the wrong one for others. This guide compares them fairly, with no prices and no vendor names, so you can match each one to the calls your hospital actually gets.

What each one is

• An IVR (interactive voice response) plays recorded menus and routes the call by the key the caller presses: “press 1 for appointments”. Some accept a few spoken words in place of keys.
• A call centre puts trained people on the phone, working from a script and a screen. It can be your own team or an outsourced one.
• An AI receptionist, also called an AI voice agent, is software that listens to what the caller says in their own words and talks back. Many older voicebots understood only a fixed list of phrases. Connected to your schedule, an AI receptionist can book, move and cancel appointments itself.

How they compare, side by side

Why IVR menus struggle in Indian OPDs

AIIMS New Delhi tried almost every option and wrote down what happened. Its white paper on the appointment system describes booking by web portal and by IVRS from 2014. The IVRS had hard limits: only patients who already had a hospital ID number could use it, they couldn’t choose a doctor, and they were booked with the doctor who had seen them before. On 14 April 2015, AIIMS started a phone call centre, because most of its patients were illiterate and “not able to utilize the web portal/IVRS”.

Then it compared who turned up. New patients who booked through the call centre came 46% of the time; those who booked through the IVRS, 23%. For follow-up visits the figures were 63% and 40%. AIIMS is a free public hospital with enormous demand, so its absolute numbers aren’t typical, and different patients use different channels. But the gap is wide.

Apps and online forms are no easy answer either. In a survey of 400 OPD patients at a tertiary hospital in Uttarakhand, 99.5% could use a smartphone and 71% knew online registration existed, yet only 26% had used it; the main barrier was not knowing how (Kalyani and others, 2021).

The difference shows on a real call. An IVR makes the caller fit its menu. Here is an example written for this guide:

The caller has to know which department they need and sit through each list. An AI receptionist lets them start with the problem, in their own words (again an example, with made-up details):

The caller never had to know the word for orthopaedics, or which key leads to it.

When an IVR is the right choice

IVRs are not the villain. They are predictable, they never get tired, and for some jobs they are the right tool:

• A handful of clear destinations, such as emergency, OPD, lab reports and billing, for callers who know where they want to go.
• After-hours messages: what to do in an emergency first, then the OPD timings and when you will call back.
• Simple information that everyone needs, such as OPD hours, played before the menu.
• A fallback for when your other systems are down.

An IVR does these jobs well when the menu has no more than three choices, the emergency message comes first, the prompts are in the language most of your callers speak, and every level lets a caller reach a person or the call-back list. It does badly when it is asked to book appointments for callers who don’t know which department they need.

What a call centre does better, and where it struggles

People are still the best answer for calls that need judgement: a family weighing up surgery, a patient with an insurance or TPA question, a complaint, a confused elderly caller, a frightened parent. A good agent can calm, explain and decide when to leave the script.

The weakness is arithmetic. Each agent takes one call at a time, and OPD calls bunch up in the morning. Even dedicated booking teams lose a lot of callers: in the UK, two public outpatient booking services abandoned 30.1% of calls, with an average wait of 5.61 minutes to be answered (NHS Lothian, 2026), and 26.1%, with an average wait of 3 minutes 55 seconds (Southern Trust, 2026). Staffing adds up too. One seat open from 8 am to 8 pm every day needs about 2.6 people once weekly offs and leave are counted; the arithmetic is in our guide to what a hospital phone desk really costs.

If you outsource

An outsourced healthcare call centre saves you hiring and training, and it can scale up quickly. Before you sign, check four things:

1. Do agents book into your live schedule, or take requests that your staff must confirm later? Requests mean every booking is handled twice.
2. What happens to an emergency caller, step by step, and who at your hospital is told?
3. Which languages are covered, and at what hours? Ask for the roster, not the brochure.
4. Who is responsible for patients’ data? Under India’s Digital Personal Data Protection Act your hospital is the data fiduciary, and the call centre is a data processor that handles data on your behalf. The DPDP Rules notified in November 2025 give organisations eighteen months to comply (PIB, 2025). Put security, deletion and breach reporting in the contract.

Where an AI receptionist fits, and where it doesn’t

An AI receptionist is built for volume: the booking, rescheduling and “डॉक्टर साहब कितने बजे बैठते हैं?” (what time does the doctor sit?) calls that fill a morning, and the calls that come in at night and on Sundays. A good one talks the way your callers talk, books only into times that are really free, and passes everything it can’t handle to your team with the reason.

It is the wrong tool for counselling a family about surgery, settling a billing dispute, handling a complaint or anything clinical. It must never diagnose or suggest medicines, and its emergency handling is a safety net, not triage: the caller is told what to do now, and your staff are alerted. Our guide to what an AI receptionist does covers those limits in detail.

DrMate’s AI receptionist answers your existing number in everyday Hindi and Hinglish, day and night, and checks every sentence a caller says for emergency warning signs before anything else. When a caller asks for a person, it can pass the call to your staff during the hours you choose, and if nobody picks up, the caller goes on your call-back list. You can test it against your current IVR on our demo line, 080 6553 5931, where it answers as Sunrise Hospital, a sample hospital we made up.

How to combine them

Most hospitals will end up with a mix, not a single winner. A sensible design:

1. Emergencies first. Whatever answers, the first seconds of the call must tell an emergency caller what to do, or recognise the emergency and act on it.
2. Routine calls to software. Bookings, changes, timings and fees go to the AI receptionist, or to a short IVR with a call-back option if that is what you have.
3. Judgement calls to people. Your reception team or call centre takes what needs a person during their hours; outside them, the call goes on a call-back list with the reason.
4. One weekly report. Count abandoned calls at every step, inside the menu included.

Questions to ask before you choose

1. What share of our calls are routine bookings and questions, and what share need judgement? A week of written call reasons will tell you.
2. When do our calls peak, and what happens to callers then?
3. Which languages do our callers speak, and in what mix?
4. Does it book into our live schedule, or only take requests?
5. What happens in the first ten seconds of an emergency call?
6. Which report shows us abandoned calls, including hang-ups inside menus?

If you have not yet counted your missed calls, start there: our guide to missed calls at hospitals shows how, and how a pilot works explains how DrMate starts with the calls your team misses and the ones after hours.