An AI receptionist that answers on a real model, not a script
Most "AI assistant" demos are a decision tree wearing a chat window. This one is wired to a live model, so you can try to break it — ask something off-script and see what happens.
The problem
A small clinic loses patients in the gap between calls. The phone rings while the front desk is with someone, the voicemail is never returned the same day, and the patient books with whoever picks up first. Nobody is doing anything wrong — there is simply one person and three things happening at once.
The approach
Give it a job, and a boundary
The interesting question with a clinic assistant is not what it can answer, but what it must refuse. This one is scoped to scheduling, services, hours, insurance and directions — and it is instructed to hand off to a human the moment a message looks clinical. It will not tell anyone whether their symptom is serious. That boundary is the design, not a limitation of it.
What’s in the build
- A live model behind the chat, so answers are generated rather than matched to a script
- Grounded in the clinic's own material — services, hours, insurance accepted, location
- Appointment capture: reason, preferred time, contact details, handed off as a structured request
- An explicit refusal path for anything clinical, with a route to a human instead
- Bilingual by default, because the front desk already is
Who this fits
- Clinics and practices where the phone goes unanswered during appointments
- Any business losing after-hours enquiries to whoever replies first
- Front desks handling the same twenty questions every day
Decisions worth explaining
Refusal is a feature
The fastest way to make a clinic assistant a liability is to let it sound confident about symptoms. Ours is instructed to stop and hand off. It is worth being unhelpful in a narrow band to be trusted in the wide one.
Scoped to what the clinic actually knows
The assistant answers from the clinic's own information, not the open internet. A wrong answer about opening hours is a lost patient; a wrong answer invented from general knowledge is worse.
About this demo
Clínica MedValle is a fictional clinic built to demonstrate the approach. The services, hours and staff are invented — the assistant is genuinely running on a live model, so its answers are generated in real time and are not scripted.
Questions
What does an AI agent like this cost?
Our AI work runs $1,200–$4,000 depending on how many systems it has to touch and how much of your material it needs to be grounded in. A scoped assistant answering questions and capturing bookings sits in the middle of that range.
Will it say something wrong to a patient?
It is scoped so the expensive mistakes are off the table: it answers from your own information rather than general knowledge, and it is instructed to hand off to a human on anything clinical. We would rather it say "let me get someone" than guess.
Can it book into the calendar we already use?
Yes. It can hand a structured request to your scheduling tool, or email it to the front desk if you would rather a person confirms every booking. Most clinics start with the second and move to the first once they trust it.
Let’s start
Let’s talk about
your business.
20-minute discovery call. No pitch, no pressure — just a conversation about what your business actually needs and whether we’re the right fit.