· Engineering · Triagents Engineering
Inside modality-aware booking
Why a generic 'book an appointment' agent can't run a radiology front desk — and what we built instead.
A generic “book an appointment” agent breaks the first time someone asks for an MRI. This post explains why, and what we built instead.
The problem in one sentence
In radiology, “the appointment” is a five-variable function of the modality, the body part, the contrast requirement, the funding pathway, and the radiologist on shift — and the patient on the phone usually only knows two of those variables.
What goes wrong with a generic agent
A generic booking agent treats the appointment as a single slot in a calendar. It hears “MRI” and offers Tuesday at 10:30. But the right answer depends on:
- Which MRI machine. A 1.5T MRI is not a 3T MRI for cardiac imaging.
- Which body part. A cervical-spine MRI takes 25 minutes; a multiparametric prostate MRI takes 60.
- Contrast or no contrast. Contrast bookings need a sub-clinician in attendance.
- Funding pathway. Bulk-billed vs. private vs. workers-compensation determines the slot pool the booking can come from.
- Referring practitioner. Some specialists require the patient to call back with a referral number; some don’t.
If the agent doesn’t ask about these, the booking is wrong. If the agent asks about all of them, the patient hangs up.
What we built
Modality-aware booking is a first-class object in TriageFlow. Each clinic defines:
- The modalities they offer (and which machine handles each)
- The body-part taxonomy and per-body-part duration
- The contrast/no-contrast rules per body part
- The funding pathways and their slot pools
- The referrer-specific intake requirements
TriageVoice and TriageChat read this configuration in real time. When a patient asks for an MRI, the agent’s next question is shaped by what it already knows from the conversation — not by a generic decision tree.
The result
Bookings made by Triagents have a measurably lower rebooking rate than bookings made by generic AI agents — because the booking captures everything reception would have asked for on a manual call. The patient leaves with the right slot the first time.
What’s next
We’re rolling the same model out to surgical scheduling next, with a different five-variable function but the same shape. If you run a surgical group, get in touch.