The problem
A Cairo dental clinic books twenty patients by phone and sees fourteen. Nobody confirms the day before. The receptionist spends her morning answering calls instead of recalling the waiting list. The doctor’s idle hour is gone money — and the patients who would have paid for it never hear the slot opened.
The build
Two people ship a WhatsApp number that acts as the clinic’s receptionist. Patients write or send voice notes in Arabic. The model sorts routine bookings from urgent ones, offers open slots from the clinic’s calendar, and writes the booking down. The evening before each appointment it asks for a yes or no. A no — or silence — frees the slot and messages the waitlist first-come-first-served. The clinic staff get a single day-sheet screen: who is coming, who cancelled, which slots refilled.
Why now
Two price drops make this a software assembly job. Since July 2025 Meta charges per template message, but the back-and-forth that matters — patient replies inside the 24-hour window — is free, as are utility confirmations sent in-window. And Flash-class models now read and transcribe Egyptian Arabic at roughly ten cents per million input tokens, with a free tier for the pilot. Three years ago the same bot would have needed a call center and a dialect speech team.
First customer
Private dental and dermatology clinics in Cairo and Giza with three to five chairs. Their pain is countable: each empty chair-hour has a price, and the owner sees it daily. They prepay a monthly per-chair fee because the bot starts recovering revenue in its first week — one saved no-show covers the month.
Challenges
Dialect voice notes will be misheard; the design must fail into a human callback, never a confident wrong booking. The bot must never diagnose — urgency wording always ends with confirming at the clinic. And patient lists are sensitive, so the pilot runs on the clinic’s own WhatsApp number with a plain written promise about where messages go.