WhatsApp for clinics: making the place the sale happens measurable
In health tourism the sale closes on WhatsApp, yet in most clinics it is recorded nowhere. Business API, the 24-hour rule, assignment and conversation analysis.
Ask where the sale closes in health tourism and the answer is almost always the same: WhatsApp. The patient sees an ad, writes, talks for days, and eventually either travels or does not.
The problem is that in most clinics those conversations are recorded nowhere. The thread on the consultant’s phone is that person’s data, not the company’s.
That has three concrete costs.
Three blind spots
When a consultant leaves, the history leaves too. Months of conversations, quotes given, promises made — all inside one phone.
The source disappears. If the patient came from a click-to-WhatsApp ad, that ad’s identity arrives with the first message. On a personal phone it is written nowhere, so nobody knows which campaign produced a patient.
Quality cannot be measured. How fast a consultant replies, how they handle an objection, whether they nailed down a next step — all of it requires someone to sit and read the threads. Nobody does.
Personal WhatsApp vs Business API
| Personal / Business app | Business API | |
|---|---|---|
| Multiple consultants, one number | No | Yes |
| History stays with the company | No | Yes |
| CRM integration | None | Yes |
| Ad source carried through | None | Yes |
| Automatic assignment and handover | No | Yes |
| Template messages | Limited | Via approved templates |
A small clinic with one consultant can manage on the app. With three consultants and a few hundred conversations a month, the API stops being a preference and becomes a requirement.
Where setups get stuck
The 24-hour rule
You cannot send free-form text more than 24 hours after the customer’s last message; only pre-approved templates are allowed.
This bites hard in health tourism, because patients say “let me think” and disappear for a week. When the window closes, the interface should disable the free-text field and offer approved templates. Do not expect consultants to memorise the rule — let the software remember it.
Prepare templates in advance: follow-up, price reminder, appointment confirmation. Approval can take days; starting after the campaign launches is too late.
Repeated delivery and ordering
The same webhook can arrive more than once, and messages may not arrive in the order they were sent. Build the conversation timeline from each message’s own timestamp rather than delivery order, and deduplicate on the message id.
Miss this detail and the patient sees a message twice, or reads the conversation out of order. Both cost trust.
Assignment and collision
Messages to one number land in one shared inbox. Two consultants writing to the same patient at once is worse than neither writing.
What you need: automatic assignment rules, handover, internal notes, and a “someone is typing” indicator.
Response time: the one number to watch
In health tourism a patient messages several clinics at once. Whoever replies first gains a large advantage.
But do not look at the average. It can read twelve minutes while a fifth of messages wait six hours — and the lost patients are precisely in that fifth.
Track:
- The 90th percentile of first response time, not the mean.
- Wait times for messages arriving at night and at weekends. Time zones make this worse with international patients.
- The number of unanswered conversations. It should be zero.
Conversation analysis
Once conversations are recorded, something previously unmeasurable becomes measurable: sales quality.
Define what gets evaluated in advance: whether qualifying questions were asked, how objections were handled, whether a next step was made concrete, where the conversation broke down.
One warning: never treat a language model’s score as a metric on its own. Keep a fixed, human-labelled evaluation set and re-measure scores against it whenever the model or the prompt changes. Low scores should not go straight onto a performance scorecard — they should enter a human review queue. The model is a filter here, not a judge.
How far to automate
Repeat questions have a high recurrence rate: price range, process, accommodation, recovery timeline, payment options. A layer answering those frees significant consultant time.
Three rules are mandatory:
- No answer without a source. Let the system answer only from your own documents; with no grounding, it should not answer.
- Hand off below the threshold. The handoff rate is not a number to hide — it is the headline metric.
- The medical domain stays closed. Eligibility, diagnosis, medication and outcome questions go straight to a person. In healthcare, a model being convincingly wrong is not an acceptable risk.
We covered how that layer is built in our AI automation guide.
The order to build in
- Move to the Business API and put the number in the company’s name.
- Connect the inbox to your CRM so the ad source lands on the contact record.
- Prepare templates and submit them for approval.
- Start measuring response time at the 90th percentile.
- Only then add the automated answer layer.
To collect enquiries in one place and track their source through to the sale, see CRM software.
Clinics that do step five first have put a system they cannot measure in front of a patient.
