AI front desk for Singapore clinics: what it should book, and what it must hand over

In this guide

• What it can do on its own
• The four hand-over cases
• Request types at a glance
• Reminders and confirmations
• No medical advice
• The PDPA and patient messages
• What to measure
• Phone receptionists and how Olano fits

What can an AI front desk do for a Singapore clinic on its own?

On its own, an AI front desk can answer the administrative questions a clinic has already written answers to (location, opening hours, parking, payment, listed prices), send reminders and confirmations for booked visits, and gather what a booking request needs: new or returning patient, the service, and preferred days and times. It answers only from the clinic's own documents, never by guessing.

Singapore's Ministry of Health draws a similar line. Its Artificial Intelligence in Healthcare Guidelines (AIHGle 2.0, March 2026) sort healthcare AI into Clinical, Clinical-Ops and Ops uses, name "AI software used to support scheduling or inventory management" as Ops, and point Ops uses to sector-agnostic guidance such as IMDA's Model AI Governance Frameworks (our explainer). Our reading, not MOH's: a booking agent stays an operational tool only while it stays out of clinical questions.

When must an AI front desk hand over to staff?

An AI front desk must hand over whenever the right answer depends on clinical judgement, on changing an existing booking, on who the patient's practitioner is, or on a confirmation only the clinic's system can give. Four cases cover most of it; symptoms get their own rule, below.

1. A clinical need is not a scheduling request

A patient asking about an implant, a crown or a tooth that hurts is expressing a need, and is not necessarily ready to be scheduled. The next free slot may be the wrong appointment type, length or practitioner, or skip the consultation the clinic wants first. The agent should collect contact details, preferred times and what the patient described in their own words, say a team member will follow up, and pass the conversation to staff, without asking diagnostic questions.

2. A reschedule is two changes, not one

Moving a booking releases one slot and secures another. If the agent cancels first and the new time is gone, the patient has no appointment at all. So the order is fixed: secure the new slot in the clinic's system, release the old one, then confirm both in one message. A booking can also carry a deposit, a lab case or the next visit in a treatment plan, so many clinics route reschedules to staff. The agent can still take the request and the times that suit.

3. The same dentist or therapist is a continuity question

A returning patient midway through treatment, or one who wants the dentist or physiotherapist they saw last time, is asking a continuity question, not a slot-filling one. The agent should offer only that practitioner's slots when it can see that calendar, and hand over when it cannot, or when the patient's records rather than the chat decide who they should see.

4. An available slot is not a confirmed booking

A free slot on a calendar is a proposal. The booking exists once the clinic's own system has accepted it, because a receptionist may have given that slot to a caller a minute earlier, or the treatment may need a longer block. Until then the agent should say requested or reserved, not booked, and send the confirmation only after the system confirms.

Request types at a glance

What the agent does with each common request:

Scroll the table sideways to see every column.

How should a clinic send appointment reminders and confirmations?

Send a confirmation once the clinic's system confirms a booking and a reminder before the visit, for example the day before; keep both factual (date, time, practitioner, address, how to change it), and on WhatsApp send them as approved utility templates whenever the patient has not written in the last 24 hours.

Outside that 24-hour window WhatsApp delivers only templates it approved in advance. For Singapore numbers, Meta charges US$0.016 for each utility template, such as a reminder, and US$0.016 for each reply inside the window after the first 1,000 free service messages per business number a month; a reminder mixed with a promotion is billed as marketing, at US$0.0732. At list rates, 1,200 reminders a month cost US$19.20 (Meta's rates as checked on 16 October 2026; sources in our WhatsApp Business API pricing guide).

Factual wording matters under the PDPA too. The PDPC's Advisory Guidelines for the Healthcare Sector (revised 20 September 2023) say a text sent solely to remind a patient of an appointment is unlikely to be a "specified message" under the Do Not Call provisions, while a message promoting a product or service is one, and needs a DNC Registry check unless the patient gave clear and unambiguous consent.

Preparation instructions, such as fasting before a procedure, are clinical content: the agent sends the clinic's approved wording exactly. And a reply of "can't make it" turns a reminder into a reschedule.

Why must an AI front desk never give medical advice?

An AI front desk cannot examine anyone, and a fluent wrong answer about a symptom can hurt a patient, so the rule is absolute: no diagnosis, no triage, no treatment recommendation and no reassurance that something is normal. AIHGle 2.0 lists hallucination (incomplete or fictitious content presented as fact) among the risks generative AI amplifies. Before launch, write down three things:

• Triggers: pain, swelling, bleeding, fever, a medication reaction, a problem after treatment, distress, or anything the patient calls urgent.
• One fixed reply the clinic writes, including what to do in an emergency, so the agent never improvises.
• A named owner who sees each escalated conversation in full, after hours included.

Promotion has rules of its own. For a clinic licensed under the Healthcare Services Act, the Healthcare Services (Advertisement) Regulations 2021 require advertising of a licensable service to be factually accurate, with no laudatory statements or superlatives, no reviews or testimonials outside narrow exceptions and no before-and-after images. An agent should not call the clinic the best in anything or forward patient reviews.

What does the PDPA mean for patient messages?

The PDPA applies to what patients send an AI front desk as it does to the clinic's other records: use it only for the purpose the patient gave it, protect it, keep it no longer than needed, and treat some health details as high-risk.

• Purpose. The PDPC's healthcare guidelines say deemed consent by conduct does not cover purposes beyond those the data was provided for: a booking enquiry is not consent to unrelated marketing.
• Retention. The PDPA sets no fixed period and the guidelines say not to keep data "just in case"; rules on keeping medical records still apply. Decide whether chat transcripts join the patient record or are deleted.
• Breaches. Under the Notification of Data Breaches Regulations 2021, a breach of a patient's name or ID number with certain health details (treatment for HIV, a sexually transmitted disease or addiction, or fertility, contraception or abortion treatment, for example) is deemed to cause significant harm and must be notified; so must any breach affecting 500 or more people.
• Attachments. Photos and voice notes can be health information, so conversation logs need the same access controls as the patient record.

Consent, the Data Protection Officer and vendor questions are in our PDPA and AI agents guide.

What should a clinic measure after launch?

Measure the no-show rate, the time to first reply, and the enquiries handed over and why, each against a baseline from the weeks before launch.

• No-show rate, week by week, with late cancellations counted separately: a reminder that turns a no-show into a refillable cancellation is progress the no-show rate hides.
• Time to first reply, split into opening hours and after hours.
• Handovers by reason: clinical need, symptoms, reschedule, continuity, unconfirmed slot, or a question the documents do not answer. Many of the last kind means the documents need work; almost none means it is time to read transcripts.

When is a phone-answering AI receptionist the better choice?

If most of your patients phone rather than message, a voice AI receptionist that answers calls is the better choice, because a WhatsApp front desk will not quiet a ringing phone. Clinics whose patients already write in get more from a messaging front desk; some will want both.

The market is active: on 1 October 2026 Singapore healthtech company Zeya Health announced it had acquired ConnectLah, a clinic automation company working over WhatsApp (e27, TechNode Global). Whatever you evaluate, voice or chat, test it on the four hand-over cases above.

How Olano fits

Olano is a Singapore AI studio that builds and runs AI front desks on messaging channels: WhatsApp on the official Business Platform, plus Telegram, email and others. It does not take phone calls; it transcribes voice notes. For a clinic the agent answers administrative questions from your documents, checks your live calendar before offering a slot, sends scheduled reminders, never gives medical advice, and hands anything clinical or sensitive to staff with the whole conversation. Outbound messages wait for approval by default, and an audit trail records every reply and approval.

A single-workflow pilot might cost S$300 to S$1,500 a month plus your review time, quoted in Singapore dollars exclusive of GST as a fixed proposal after a 30-minute consultation, never per seat. See WhatsApp AI for clinics, the AI booking assistant and our WhatsApp AI agent page.