Work · Case study
Every referral routedthe first time, not the third.
01 / Challenge
Referrals were routed by whoever opened them. The same case reached the wrong department more than once, and the delay landed on the patient rather than on the queue that caused it.
02 / Approach
We trained a routing model on the group's own referral history and put it inside the existing workflow rather than beside it. A suggested destination arrives with the referral, and a coordinator confirms or overrides it in one click.
Every override is a training signal. Drift is monitored weekly, and the model rolls back to its last good version automatically if accuracy falls below the agreed floor.
03 / Outcome
Fewer re-routes
Faster to first contact
Most referrals reach the right department on the first hop, and the ones that do not are the cases a human was always going to have to look at.