The 9pm enquiry, and where the line sits
A prospective patient messages a clinic's WhatsApp at 9pm asking about a treatment — availability, roughly what it costs, what happens on the first visit. The AI reads the enquiry, checks it against the clinic's actual treatment list, and drafts a reply answering exactly that factual question. It does not draft anything that promises an outcome, describes a specific result, or reads as clinical guidance — because those are the categories KKM's advertising guidelines and, for aesthetic and dental specifically, MDC rules restrict. A staff member reviews the draft and sends it. The AI's job stops at "here's the factual answer"; it never crosses into "here's what will happen to you." This is one specific application of the wider draft-and-approve pattern covered in our AI automation overview.
Why this is a compliance-shaped build, not a generic chatbot template
Most chatbot platforms are built for e-commerce or general customer service, where an overconfident, slightly-too-promotional reply is a minor issue. For a Malaysian clinic, the same overconfident reply — "this treatment will definitely fix your concern" — is an advertising compliance problem under KKM rules, not just a tone issue. The build has to constrain what the AI is allowed to generate at the prompt and guardrail level, not just hope the model behaves. This is the same discipline our team applies across aesthetic clinic marketing and dental clinic marketing creative review — the chatbot needs the same compliance lens as an ad.
What the flow actually does
- Reads the enquiry and classifies it: a factual question (availability, price range, process), a clinical question (specific to the patient's condition), or something requiring a human immediately (a complaint, an emergency).
- Drafts a factual-only reply for the first category, matched against the clinic's real treatment list and current availability — never inventing a treatment the clinic doesn't offer or a price it doesn't charge.
- Routes clinical questions and anything sensitive straight to a staff member with no AI-drafted reply at all — the AI's restraint here matters as much as its drafting ability elsewhere.
- Logs every enquiry against the lead in the CRM, so front-desk staff see the full conversation context, not just the final approved message.
The recall and no-show layer
Enquiry triage is the AI-facing half of this system. The other half — recall lists for patients due a follow-up, and no-show reduction through WhatsApp reminders — is a booking-system and messaging problem, not an AI-drafting one. See our custom software service for how clinic booking systems are built around recall lists, and our WhatsApp automation service for the official-API reminder flows that connect to them.
Where this fails
It fails when a clinic expects the chatbot to answer genuinely clinical questions — anything specific to a patient's own condition or suitability for a treatment needs a real consultation, not an AI-drafted answer, regardless of how confidently a model can generate one. It also fails as a substitute for reviewing advertising copy for KKM/MDC compliance separately; a compliant chatbot and compliant ad creative are two different disciplines that both need attention. See our clinic advertising compliance hub for the platform-and-regulator-specific rules this connects to, and our KKM ad checker tool for reviewing creative copy directly.
What we do differently
We build the compliance boundary into the AI's prompt and guardrails at construction time — restricting what categories of question it's allowed to answer at all — rather than trusting a general-purpose chatbot platform to stay on the right side of KKM/MDC rules by default. See our AI automation service for the build process across clinics and other verticals.
What to do about it
- List the factual questions your front desk answers every day — availability, price range, first-visit process — as the chatbot's actual scope.
- Explicitly exclude clinical questions from the chatbot's allowed responses; route them to a human with no AI draft at all.
- Review chatbot-drafted replies against the same KKM/MDC standard you'd apply to ad copy.
- Connect the enquiry-triage layer to a recall and no-show reminder system so the automation covers the full patient lifecycle, not just first contact.