Overview
Accurx Scribe is an AI tool that listens to a hospital appointment and writes the clinic letter. Doctors then send it to the patient's GP, to the patient and to their medical record.
Once a letter was sent, nobody could see whether it had arrived. A letter that never reaches a GP is a patient safety risk. Hospitals were waiting, and I had one week to design a fix.
The Problem
How might we show hospital teams, at a glance, which letters still need their attention?
TL;DR
Led design end to end, from research to handover, for a new product on web, iOS and Android
Designed it in 1 week, after research with 6 clinicians showed my first designs didn't work
Tested our riskiest feature in 2 weeks, then removed it so it didn't block launch
Built 2 features cut from scope myself, using AI
My contribution
Product design
User research
Prototyping with AI
The team
1 Product Manager
1 Product Designer (me)
3 engineers
Year
2026
Process
Found the real problem by watching people work
My first six concepts grouped letters under each patient. In sessions with six clinicians, none could tell at a glance what still needed them. Asked 'Can you tell right away if it needs something from you?', a consultant said 'No.'
Rebuilt the page around letters, not patients
A patient's care is never 'done', but a letter can be. Listing letters, each with its own status, gave every row a simple answer: is this finished? I showed leadership a working prototype I'd built with AI, and they backed the change.
Designed the full product for web and mobile
A table showing where each letter went, a side panel for checking details without leaving the list, and the same flow on iOS and Android. I designed every error and empty state up front, so engineers could build it in three weeks.
Shipped a risky feature on purpose, then pulled it
We launched with a manual 'Complete' button, knowing people might not use it, and agreed up front how we'd measure it. In the first two weeks almost nobody pressed it, because they'd already done the work. I removed it so launch wasn't blocked, and planned research for a better fix.
Outcome
Results
Shipped on time on web, iOS and Android
Clinicians spotted failed letters without being told where to look
Our riskiest assumption was tested and answered in two weeks, not months
Two features cut from scope were built by me with AI, and are now live
What I took forward
If iterating isn't working, change the structure
Ship known risks with a way to measure them
When a feature isn't used, ask why before redesigning it