Case study · Healthcare
Four minutes back, every consultation.
Helios's clinicians were spending more of each appointment on documentation than on the patient. We watched, then rebuilt the three screens that mattered.
- 2,400 clinicians
- 41% less documentation time
- WCAG 2.2 AA
- Sector
- Healthcare provider
- Duration
- 7 months
- Squad
- 2 designers, 5 engineers
- Users
- 2,400 clinicians
Overview
The situation
Helios had bought a well-reviewed clinical system and rolled it out across fourteen sites. Adoption was universal, because it had to be, and satisfaction was the lowest of any system they operated.
The internal assumption was that clinicians disliked change. Two weeks of observation showed something more specific: the three most-repeated tasks each took between nine and fourteen interactions.
Challenge
The challenge
Clinical software is used in short, interrupted bursts. A flow that assumes uninterrupted attention loses the user's place every time a colleague walks in — and in a clinic, someone always does.
The system also could not be replaced. It held the record of truth and was integrated with a dozen downstream services.
- Documentation consumed 38% of a typical consultation
- Clinicians kept paper notes as a workaround, creating a second record
- The underlying system could not be replaced, only fronted
- Any change needed clinical safety sign-off before release
Approach
What we did
We spent two weeks in clinics before designing anything, then rebuilt only the surfaces where the observation showed real cost.
01
Observe the actual routine
Sixty hours of clinical observation across four specialties, timing the repeated tasks. The finding was blunt: three screens accounted for most of the lost time.
02
Design for interruption
Every flow made resumable, with state preserved on the server and a visible re-entry point. Nothing depends on the clinician remembering where they were.
03
Front the system, do not replace it
A FHIR-based service layer over the existing record, so the new interface wrote to the same source of truth with full audit continuity.
04
Draft, then have a human decide
A retrieval-grounded assistant drafts the structured note from the consultation record. It never writes to the record itself — a clinician edits and commits, and the edit distance is measured as the quality signal.
05
Evidence the safety case
The evaluation suite, known limitations and oversight design were produced as build artefacts, so clinical safety review had documents to assess rather than assurances to accept.
06
Roll out by specialty
One specialty at a time, with the previous interface available throughout. Fallback usage dropped to under 2% within three weeks of each cohort's launch.
Results
The outcome
Documentation time fell by 41% against the observed baseline. The paper workaround disappeared within a quarter, which resolved a records-governance risk Helios had been carrying separately.
Less documentation time
Against the observed baseline
Returned per consultation
Median across specialties
Fallback to old interface
Three weeks after each launch
Clinicians migrated
Across fourteen sites
“They spent two weeks watching us work before they drew a single screen. That is why the result feels like it was designed by someone who had actually been in a clinic.”
Industries
Technologies
Published 3 February 2026
More work
Other systems we have shipped
Have a problem shaped like this one?
Bring us the constraint your last partner called impossible. We will tell you on the call whether we believe it.