How Triagents handles emergencies

· Research · Triagents Team

How Triagents handles emergencies

Emergency escalation is configured by each practice — what never changes is that the AI recognises the moment, acts instantly, and gets out of the way.

The single most important safety behaviour on Triagents is what happens when a patient describes an emergency. Here is the part people find surprising: there is no single “right” response — because the right response depends entirely on the practice.

A GP clinic with a nurse at the front desk at 2pm wants that call in a human’s hands immediately. A radiology group whose phones are answered at 9pm by no one wants the caller told, clearly and without delay, to hang up and call 000. A telehealth practice may want its on-call clinician conferenced in. These are clinical-governance decisions, and they belong to the practice — not to us.

So the escalation path is configured per client. What is not configurable is the part that matters most: the AI recognises the moment, acts on the practice’s chosen path within 1–2 seconds, and stops talking.

What “emergency” means

Each practice configures its own emergency vocabulary in TriageFlow. The defaults cover a clinically broad set — chest pain, severe bleeding, stroke symptoms, severe allergic reaction, mental-health crisis vocabulary, paediatric emergency words — and the practice can add specialty-specific terms.

For radiology, the default list also includes contrast-reaction vocabulary, post-procedure complication terms, and a few imaging-specific clinical phrases that don’t show up in generic medical NLP corpora.

The response is the practice’s decision

During onboarding, each practice defines its escalation playbook — typically split by hours and staffing:

  • During business hours — most practices choose a live transfer: the call is handed to reception or a nominated clinician within seconds, with the full transcript pushed to their screen so the patient never repeats themselves.
  • After hours, or when no one can answer — the agent tells the caller plainly to hang up and call 000 (or the practice’s nominated emergency number), repeats it once, and ends its own involvement. No hold music, no “please stay on the line”.
  • Split and layered rules — business hours to reception, after hours to a 000 instruction plus an alert to the on-call phone; different rules per site, per line, or per modality. Whatever the practice’s clinical governance requires, the routing follows it.

A suburban GP clinic, a multi-site imaging group, and a telehealth-first practice will configure three different playbooks — and all three are correct, because each one was decided by the people clinically responsible for those patients.

What happens in those 1–2 seconds

When the practice’s chosen path is a live transfer:

  1. The intent classifier flags the utterance as emergency_handoff at high confidence.
  2. The audio stream is muted to the AI and the operator path is opened in parallel.
  3. The reception phone rings with a “Code Red” tag and the patient is on the line within the second ring.
  4. The full transcript and the clinical context are pushed to reception’s screen the moment the call connects — no asking the patient to repeat themselves.
  5. The event is logged with a permanent audit record.

When the path is an emergency-services instruction, the same recognition fires, the instruction is delivered immediately, and the practice is notified with the transcript and a recording reference — same audit record, same 24-hour review.

How we validate it

Three layers:

  • Per-release regression suite. Several hundred red-team utterances, including clinically obscure phrasings, sarcasm, indirect statements, and adversarial inputs designed to evade simple keyword matchers.
  • Per-practice configuration validation. Before any practice goes live, we run their configured vocabulary and their chosen escalation paths against a sample of real call transcripts to confirm the routing behaves exactly as their playbook says it should — including the after-hours branches.
  • Live monitoring. Every emergency escalation is reviewed by a clinical engineer within 24 hours of occurrence.

What we don’t do

We don’t try to triage the emergency itself. We don’t try to give clinical advice. We don’t second-guess the practice’s playbook mid-call. And we never leave a caller without a path: whatever the configuration, an emergency always ends in a human — the practice’s staff on the line, or emergency services on the way.

That’s the floor. Nothing about Triagents exists if that floor isn’t held.

← Back to blog