01 the business
Specialist hospital outpatient department
2,500 referrals a month
Referrals arrive by e-referral, secure email, fax and letter. Admin staff read each one, chase missing details and work out where it should go. Incomplete referrals bounce back and forth, and urgent ones can sit in the pile.
02 before
The problem
Referrals arrive by fax, email, e-referral and letter, in every format. Admin staff read each one to work out the service, check for missing details and chase referrers. Incomplete referrals bounce back and forth, and urgent ones can sit in the pile.
- Referrals take days to be logged.
- Missing information is found late.
- Urgent referrals depend on someone spotting them.
03 the build
What the AI does
- Step 1: Picks up referrals from e-referral systems, secure email, fax-to-email and scanned letters.
- Step 2: Reads the patient details, reason for referral, referrer and attachments.
- Step 3: Checks the referral against the service's acceptance criteria and lists what is missing.
- Step 4: Suggests the right clinic or service, and highlights words that may signal urgency for a clinician to review.
- Step 5: Creates the referral in the patient administration system and requests missing information from the referrer.
paused: waiting for a person to approve
How it was put together
- Picked up referrals from every channel, including scanned letters.
- Read patient details, reason and attachments.
- Checked against each service's acceptance criteria and listed what was missing.
- Suggested the right service and highlighted possible urgency words for clinical review.
04 human in the loop
Where a person stays in charge
A clinician sets and confirms every referral's priority; the tool only highlights possible urgency. Admin staff confirm the record before it is created.
- A clinician confirms every referral's priority.
- Admin staff confirm the record before it is created.
- The tool highlights possible urgency; it never decides it.
05 rollout
From kickoff to live in 6–8 weeks
- Weeks 1–3Agree acceptance criteria per service and channel set-up.
- Weeks 4–5Run in parallel with the admin team.
- Weeks 6–8Switch to review-and-confirm.
06 after
What changes
Referrals are logged soon after they arrive, complete ones go straight to the right clinic and urgent ones reach a clinician's eye faster.
Assume 2,500 referrals a month at 12 minutes each to read, check and log by hand, falling to 4 minutes to confirm.
2,500 × 8 minutes = 20,000 minutes, or about 333 admin hours a month.
What to measure
- Days from referral received to logged.
- Referrals returned for missing information.
- Time to clinical triage for flagged referrals.
- Admin time per referral.
07 evidence
What the research says
Under FDA guidance, software meant to support time-critical clinical decisions does not meet the definition of non-device clinical decision support.
Across 19 studies, between 6.8% and 62% of laboratory tests and between 1.0% and 35.7% of radiology tests were not followed up, with harms including missed cancer diagnoses.
08 lessons
Mistakes to avoid
- Letting software set priority. Clinicians triage.
- Poor scans. Ask referrers to use electronic channels where possible.
- Vague criteria. Write them down per service.
09 questions
Common questions
- Is highlighting urgency a medical device?
- Highlighting words for a clinician to review, without making the decision, is designed to stay on the support side. FDA guidance says software for time-critical decisions does not qualify as non-device decision support, which is why priority stays with clinicians.
- Does it work with NHS e-Referral?
- Yes, as well as secure email and scanned letters.