What may a medical practice automate?
The administration, which is most of the phone traffic. Booking and moving appointments, sending reminders, confirming that the practice takes a particular medical aid, giving the address and the parking arrangements, sending the new-patient form ahead of the visit. None of it needs a clinician, and all of it happens at hours when the practice is closed and the patient is awake.
What must it never do?
Triage or advise. A chatbot must not ask what the symptoms are and decide how urgent they sound; it must not say whether a rash needs a doctor; it must not suggest a medicine. The safe behaviour is a fixed script: “I can’t help with medical questions. If this is an emergency call 10177. Otherwise, the first slot I can book is 08:30 tomorrow, or I can pass a message to the nurse.” A chatbot trained only on the practice’s own pages stays inside that fence because it has nothing else to answer from.
| Request | Chatbot | Clinician or staff |
|---|---|---|
| Book, move or cancel an appointment | Yes | |
| “Do you take Bonitas?” | Yes, from the practice’s list | |
| Send the reminder | Yes, automatically | |
| “Is this rash serious?” | Refuses; offers a slot or the emergency number | Yes |
| Repeat prescriptions | Takes the request; never approves it | Yes |
| Results | Never in the chat | Yes |
Why do reminders matter so much in healthcare?
Because the evidence is medical evidence. The 2018 systematic review of 105 studies of appointment scheduling put the average no-show rate at about 23% and identified long lead times and a prior no-show as the strongest predictors 1. The Cochrane review of eight randomised trials with 6,615 participants found that mobile text-message reminders increased attendance compared with no reminder and with postal reminders, matched a phone call, and cost less per attended appointment 2. A practice that reminds by having a receptionist phone the list is paying more for the same result and losing the slots she does not get to.
What are the POPIA rules for patient data?
Health information is special personal information under section 26 of POPIA, which means the bar for processing it is higher than for a name and a number. The Act has applied in full since 1 July 2021; the Information Regulator can impose administrative fines of up to ten million rand, and non-compliance with an enforcement notice is an offence carrying up to ten years 3. The first fine, five million rand, was issued against a government department in 2023 4.
The practical design rules: the chatbot collects a name, a number and a reason category (“check-up”, “follow-up”), never a description of symptoms; it never asks for an ID number or a medical-aid number in the chat; transcripts are stored where the practice can delete them; results are never discussed; and anything that looks clinical is handed to a person at once.
Does a fast reply matter for a practice?
Yes, because patients choose. Harvard Business Review’s audit of 2,241 companies found firms that responded to a web enquiry within an hour were nearly seven times as likely to qualify it as firms that waited even an hour longer, and 23% never responded 5. The patient who searched for a GP on Sunday evening and got an answer and a slot has been booked; the one who filled in a form has been lost to the practice that answered.
What does the practice need to start?
Its pages, its medical-aid list, its hours and its booking rules. The booking system writes into the diary and sends the reminders; the website chatbot answers the admin questions. Both are month to month and neither says a word about medicine.
Questions people ask
Can the chatbot tell patients their results?
No. Results are never discussed in the chat; the chatbot books the follow-up and a clinician gives the result.
Does it work with our practice management software?
It books into a shared calendar or connects to your system through its integrations. Ask us about the one you run.
Is the conversation data stored in South Africa?
The chatbot’s records are held on our systems and are yours to export or delete from the dashboard; ask us for the current data-processing terms before you sign.
Sources
- Dantas, Fleck, Cyrino Oliveira & Hamacher, “No-shows in appointment scheduling – a systematic literature review”, Health Policy 122(4), 2018 (105 studies)
- Gurol-Urganci et al., “Mobile phone messaging reminders for attendance at healthcare appointments”, Cochrane Database of Systematic Reviews, 2013 (8 trials, 6,615 participants)
- Protection of Personal Information Act 4 of 2013, section 109 (administrative fines) — popia.co.za
- Bowmans, “Information Regulator issues first fine of ZAR 5 million under POPIA” (2023)
- Oldroyd, McElheran & Elkington, “The Short Life of Online Sales Leads”, Harvard Business Review, March 2011 (audit of 2,241 US companies)
Where we do this
The same build runs for businesses in Paarl, Rustenburg and Ekurhuleni, and the city pages set out what changes locally — the languages, the trading hours and the sectors that dominate. Everything is deployed and supported remotely from Vanderbijlpark.