- WhatsApp should be an interface to live scheduling, not a separate booking database.
- Natural-language time preferences need to become deterministic schedule constraints.
- The agent must revalidate the selected time before booking.
- Identity, medical boundaries and human takeover still matter even in a simple booking flow.
WhatsApp is the channel, not the source of truth
A clinic can easily build a nice WhatsApp conversation that collects a preferred day and then sends the request to reception. That is lead capture, not booking.
Real booking means the same appointment appears in the same schedule staff use, follows the same duration and provider rules, and is protected from conflicts.
Patients describe time like humans, not calendar widgets
"Next week in the morning," "after work," "anything earlier," and "the same doctor as last time" all need interpretation. The AI is very good at this part.
After interpretation, those preferences should become explicit filters for the scheduling engine. The model should not simply invent a plausible-looking time.
From vague preference to valid slots
I need something next week, preferably after 4.
Is this for a cardiology consultation?
Yes.
I can offer Tuesday at 16:30, Wednesday at 18:00 or Thursday at 17:15.
The hard part is often what happens when the answer is no
If the requested provider is unavailable, the agent needs clinic-specific rules. Can it offer another provider who performs the same service? Should it first offer another day with the requested doctor? Is provider substitution allowed for this service at all?
These are business rules, not language-model preferences.
The selected slot must be checked again before booking
A conversation can take several minutes. Another patient may take the slot in the meantime. The booking operation should therefore validate the exact requested time at commit, not trust an earlier suggestion.
If the slot is gone, the system should say so and offer fresh alternatives. It should never force the booking through because the assistant promised it.
Booking is only one part of the appointment lifecycle
Patients also need to confirm, move and cancel existing appointments. Those actions require stronger identity and state checks than showing public availability.
The conversation should identify which appointment is being changed, make the effect explicit and confirm the final state after the application completes the operation.
The booking flow will eventually collide with a medical question
"Do I need a referral?" may be an administrative question if the clinic has an approved rule. "Can I wait until Thursday with these symptoms?" is not.
The patient should not need to understand the boundary. The product should detect that the conversation crossed it and escalate appropriately.
How Menaela treats WhatsApp booking
Menaela treats WhatsApp as one interface to the clinic. The conversation uses the same provider/service relationships and scheduling engine as the product UI, and the final mutation is validated by the application.
That keeps the convenience of chat without creating a parallel appointment universe.
