An NHS acute trust running seven specialty pathways
Referral triage, manual to
automated in twelve weeks
2,400 referrals a month, DSPT-compliant throughout, and a clinician signature on every single score.
- Sector
- NHS · acute trust
- Engagement
- Production deployment
- Timeline
- 12 weeks, including IG sign-off
- Compliance
- DSPT · UK GDPR · CQC
The challenge
What the team was living with
The trust’s referral management team was processing every incoming referral by hand — reading, categorising and assigning clinical urgency across seven specialty pathways. At 2,400 referrals a month with a growing backlog, the trust was at real risk of breaching 18-week RTT pathway requirements.
Clinicians were spending up to 40% of their administrative capacity on triage alone. The referral letters themselves were unstructured and of highly variable quality, with missing data a routine rather than exceptional problem.
Every one of those touchpoints carried DSPT, UK GDPR and CQC obligations, and NHS procurement constraints limited how quickly anything could be deployed at all.
- 2,400 referrals a month processed entirely by hand
- 18-week RTT pathway breach risk from the triage backlog
- Inconsistent urgency scoring across the team
- Unstructured referral letters — variable quality, frequently missing data
- NHS procurement constraints limiting speed to deploy
- Strict DSPT, GDPR and CQC obligations on every patient data touchpoint
The approach
Augment the clinician’s judgement.
Never replace it
Establish the information governance path first
The DPA, the DSPT position and the IG approval route were agreed before architecture began. On NHS work this sequencing is the difference between twelve weeks and twelve months, and it is the step most vendors leave until last.
Deploy inside the approved boundary
The system was deployed on NHS-approved Azure HSCN infrastructure, with patient data never crossing the approved boundary at any point in the pipeline.
Extract structured clinical data from unstructured letters
Referral letters are read and structured clinical data extracted, with missing fields flagged explicitly rather than inferred. An inferred clinical field is a patient safety issue, not a convenience.
Score urgency across seven pathways, with reasoning
Clinical urgency is scored per pathway with an XAI reasoning trace attached, and a red-flagging system escalates automatically on urgent clinical indicators.
Make human review mandatory, not optional
A clinician reviews every AI score before anything routes. The system ranks; the clinician decides. Every override is captured and retained with the record.
Build the audit trail to CQC expectations
Every triage decision, its AI reasoning and its clinical sign-off are logged together, in the form an inspection actually asks to follow.
The architecture
Governance designed in, not bolted on
Every architectural decision on this engagement was made with NHS information governance at the centre. The compliance layer is not a wrapper around the system — it is a constraint on every layer of it.
- NHS-approved Azure HSCN
- Data never leaves the boundary
- UK GDPR Data Processing Agreement
- Referral letter intake
- Structured clinical extraction
- Missing-data flags raised explicitly
- Urgency scoring × 7 pathways
- XAI reasoning trace
- Red-flag auto-escalation
- Mandatory clinician review
- Override capture
- No routing without sign-off
- CQC-compliant audit log
- Clinical rationale retained
- DSPT evidence pack
Outcomes
What changed, measured
No clinical decision moved away from a clinician. What moved was the reading, sorting and structuring that came before the decision — which is where the forty per cent of administrative capacity was going.
Take it with you
The full engagement, in twelve pages
Architecture decisions, what we got wrong on the way, and the measurement methodology behind every figure on this page.