What was in the way
Mersey Health Group processes thousands of inbound referrals across its clinics. Triage was manual: clinical administrators read every letter, classified urgency, and routed each case to the right pathway.
The group had attempted AI three times before and each pilot stalled. They were impressive demos that never survived contact with real data, governance review or day to day load.
What we built
We treated this as a production engineering problem from day one. After a focused strategy and risk review, we built a document intelligence pipeline that reads referral letters and extracts structured, reviewable data.
On top of it, a triage assistant proposes an urgency band and pathway, always grounded in the source document and always cited. Nothing is auto actioned, and a clinician confirms every decision.
We hardened the system with evaluation harnesses, audit logging and clear governance guardrails, then trained the administrative team to run and monitor it themselves.
By the numbers
For the first time, the group has an AI system in daily clinical use. Referrals reach the right pathway faster, and every suggestion is traceable back to the source letter.
“We had three failed AI pilots before this. The difference was treating it as a production engineering problem from the start.”