Skip to content
Modern hospital room with digital medical equipment

CS-02 · Healthcare

Referral triage for a regional health network

Faxed and emailed referrals were sorted by hand before a clinician could prioritise them. An intake agent now reads each referral, extracts the essentials and queues it for clinical review with nothing decided on its own.

Referral to first clinical review
2 days → same day
Fields extracted per referral
11
Referrals reviewed by a clinician
100%
Discovery to production
10 wks

The situation

The bottleneck was the inbox, not the clinicians.

Referrals arrived as faxes, PDFs and emails in every format imaginable. Administrative staff opened each one, keyed the details into the patient administration system and passed it on for clinical triage.

On a busy week the backlog reached two days before a clinician saw a referral. Nobody wanted a model making clinical decisions; everybody wanted the keying to go away.

What the agent may and may not do

ActionAgentPerson
Read the referralYesYes
Extract patient and referrer detailsDraftsConfirms
Suggest a specialty queueSuggestsDecides
Assign clinical urgencyNeverClinician only
Contact the patientNeverYes

Guardrails the clinical governance committee asked for

Rule 1

Nothing is decided by the agent

It prepares; a person confirms every field and a clinician sets every urgency.

Rule 2

Uncertainty is visible

Fields the agent is unsure of are flagged, not guessed.

Rule 3

Everything is logged

Each extraction keeps the source page and the reviewer who confirmed it.

All case studies

Drowning in documents that all need keying?

We'll look at a sample with you and say plainly what can be automated safely.