AI Receptionist for Medbase Clinics

The most expensive thing a clinic's phone does is invisible. A call rings out at lunchtime, a mailbox fills overnight, a caller hits an engaged tone at nine on Monday and rings the practice down the road instead. None of it appears on a report, because nothing malfunctioned: the phone worked, it just was not answered. A clinic on Medbase can have a spotless system and a good diary and still lose a steady trickle of patients to the calls it never picked up.

We build the Medbase connection, and this is about the calls a report never shows, and what happens when every one of them is answered and written into Medbase instead.

The reason missed calls are so easy to underrate is that they leave no trace. There is no error log for a caller who gave up, no line item for the booking that never happened. The cost is real and the evidence is absent, which is the worst combination for a problem to have.

When a call is missed	What the practice sees	What actually happened
It rings out at lunch	Nothing on any report	The caller tried the next clinic
Voicemail after hours	A full mailbox by morning	Most callers left no message
An engaged tone at peak	No record at all	A booking lost to a busy signal
A long hold, then a hang-up	No record at all	An abandoned call, simply gone

Answering every call is the whole point, because the only way to stop losing the invisible ones is to make sure none of them is missed in the first place. There is no per-line limit, so a hundred callers at once are each answered on the first ring, and the mailbox that used to fill overnight is replaced by a morning of appointments already booked.

The overflow the desk cannot cover

Missed calls are not only an after-hours problem. The busiest hour of a weekday is when a clinic is most likely to lose one, because that is exactly when the desk is already on another line, checking a patient in, or three deep in a queue. The calls lost at 9am on Monday are the same as those lost at 9pm, just harder to see, because a receptionist is visibly working the whole time.

2care sits in front of the same line and takes the overflow the instant the desk cannot. It answers the second and third simultaneous call a single receptionist physically cannot, books them, and hands over only the ones that need a person. The desk is not replaced; it is stopped from being the bottleneck at the hour the clinic can least afford one.

How a Medbase booking is made, and how fast

Underneath, one mechanism handles all of it. 2care reaches Medbase through its API, authorised with a scoped credential set during onboarding, with nothing installed at the clinic. Availability is read live by provider and appointment type, so an offered slot genuinely exists, and the booking is written against the resolved patient and validated by Medbase against its own rules, so a slot taken during the call fails cleanly rather than being assumed free.

The speed is why a caller stays on instead of hanging up. A reply lands in about 480 milliseconds, ninety-fifth percentile under 700, on a voice pipeline built in-house rather than assembled from outside parts, so reading Medbase live is not a silence the caller notices. Uptime holds at 99.9 per cent across a thousand and more simultaneous calls, and the booking write is idempotent, so a retry after a dropped connection creates one appointment, not two. Turn-taking waits for a complete thought rather than the first pause, so a caller who hesitates mid-sentence is not cut off.

From a reason to the right appointment

Callers do not speak in appointment types; they say why they are ringing. 2care resolves the reason, provider and urgency from natural speech and maps it to the Medbase appointment type the clinician offers, at the right length, so a caller never has to translate their problem into a menu option they cannot map. A request that does not fit a clinician's rules is offered an allowed alternative rather than written and bounced, and anything that turns clinical leaves the booking flow and reaches a person, configured in triage and routing.

A refill or a results question is handled without taking a scheduling slot: the refill is captured with the drug and pharmacy and routed to the provider queue, the results question is routed to the clinical team with the patient already resolved. Keeping those off the booking line is part of why the calls that do need a slot get answered faster.

Take a caller who says, "I just need to see someone, I've had this cough for about ten days and it's not shifting." There is no menu option for that, and a phone tree would strand them. The agent recognises a returning patient, reads the cough against the escalation triggers, and if it stays routine, offers the right appointment type and length with an available clinician. The caller never had to know whether that was a standard consult or something longer; the clinic's own rules in Medbase decided it, and the booking reflects them.

The record, resolved before the write

Before anything is written, the caller is matched to the existing Medbase record and confirmed on a date of birth, so a returning patient keeps one chart; a genuinely ambiguous match is passed to a person with the candidates gathered, using patient matching, rather than a guess that splits a history across two charts. Intake is captured in the same breath and written to Medbase as structured fields, not a note someone retypes later, and the same capture feeds patient intake whether the patient rang, messaged or filled in a form. Identity is settled before a slot is ever held, so a detail never lands on the wrong chart and no appointment is reserved against a patient the agent has not actually confirmed.

What 2care changes for a Medbase clinic

A Medbase clinic is not short of a system that can hold a diary and a chart. It is short of a phone that answers every call and turns it into a clean booking without another receptionist. 2care does exactly that: it answers, resolves the reason, books against Medbase's own availability and rules, resolves the patient, and writes the result back, in one native pipeline rather than a chain of tools passing a patient along.

What a clinic feels is not a faster booking on the calls it already caught. It is the recovery of the ones it never did: the lunchtime rings, the evening voicemails, the Monday busy tones, turned back into appointments. On a business whose costs barely move whether a slot sells or not, the calls answered that used to be lost are close to pure margin, and they are the ones a report was never able to show the clinic it was losing.

The effect compounds in a way a single month understates. A caller who could not get through once tries elsewhere; a caller who books becomes a patient who returns and refers. The difference between answering nine calls in ten and answering all ten is not one appointment, it is the stream of future visits attached to the patient that would otherwise have been lost, which is why clinics tend to notice the change a quarter in rather than a week in.

Where the data goes, and under what agreement

A booking on this line is not a call the words pass through and leave nothing. It builds a transcript, matches a patient, and writes an appointment into Medbase, which is creating and holding health information rather than merely carrying it.

Under HHS guidance and the equivalent duties elsewhere, that makes the vendor a business associate rather than a conduit. So a signed agreement covers the connection before the first live call, the data is encrypted in transit and at rest and held in region where the law requires, no model behind the conversation trains on it, and every write is reversible. The compliance page sets out the detail on each side.

Where this is right for Medbase, and getting live

This is right for a clinic that loses calls it cannot see, keeps its providers, appointment types and rules current in Medbase, and has a clinician to take the calls that turn clinical. Given those, the benefit lands from the first day the line goes live, and the quietest evidence of it is a voicemail box that is no longer full in the morning.

Getting there is three moves over weeks. The connection is authorised to read availability and write appointments through the Medbase API. The clinic's setup is mapped: providers, appointment types and durations, and the route for a clinical call. Then the line is tested against a sandbox with the team listening before it is switched across, and the routing is tuned over the first days against the real call mix, each change backed by the trigger every escalation records rather than a hunch.

The tuning window is short because the evidence is concrete. An appointment type mapped to the wrong length is corrected once and stays corrected; a trigger firing on too many ordinary calls is widened against the recorded examples, not a guess. Within a fortnight the routing reflects how the clinic actually runs, and the number patients dial has not changed at any point, because the agent answers in front of the existing phone system rather than replacing it. There is nothing to port and no hardware to install.

Frequently asked questions

Does it book real Medbase appointments or take a message?

Real appointments, written through the Medbase API against the patient's chart, and read back to the caller only after Medbase accepts the write. On the rare call it hands to a person, the full request travels too, so nothing is lost in the handover.

What happens to calls after hours?

They are answered, every hour, at any volume, with no busy signal. Out of hours the agent books into your next open slots and follows your escalation rules for anything that cannot wait, so the morning starts with appointments made rather than a mailbox to work through.

Will it create duplicate charts in Medbase?

No. The caller is matched to their existing chart first and confirmed on a date of birth, and where two charts are equally likely a person takes over rather than the agent guessing, so a history stays on one record.

Can it handle refills and results questions?

Yes, without taking a scheduling slot. A refill is captured with the drug and pharmacy and routed to the provider queue; a results question is routed to the clinical team with the patient resolved. Keeping both off the booking line frees it for the calls that need a slot.

Is patient data ever used to train models?

No. Nothing a patient shares becomes training data for any model in the pipeline. It is encrypted end to end, held in region, kept only as long as the record requires, and covered by the agreement that governs every Medbase call the agent takes.

Hear it answer the calls you never see

The measure of a Medbase line is not the calls it books that a receptionist would have booked anyway. It is the lunchtime and after-hours calls that used to ring out, answered and turned into appointments before anyone at the clinic knew they had come in.

See how the same connection runs on other systems across integrations, or hear the calls a report never shows handled on a live line when you book a demo.