My Framer Site

HEALTHCARE AI

Smart Notes: AI-powered patient summaries for oncologists

Oncologists at Mayo Clinic spent 15+ minutes per patient reading dense clinical notes before each consultation. Our two-person team rebuilt that summary into an AI-assisted Smart Note and a longitudinal timeline. In a formative study with practicing oncologists, review time roughly halved.

Role

UX Designer

Team

2 designers

Timeline

Jan - Apr 2025

Tools

FIgma, FigJam

RESEARCH

Designed with oncologists, not just for them

Three methods, two oncologists, three patient scenarios — new diagnosis, follow-up, and treatment change.

Method 01

Day-long shadowing

Contextual inquiry in clinic, watching how oncologists parsed current notes between consults — where they paused, scrolled back, re-read.

OUTPUT

Workflow map + 10 friction points

Method 02

Co-ideation sessions

Sketched directions alongside the oncologists, pressure-testing what would actually work in the flow of a consult before we committed to prototyping.

OUTPUT

Clear insights to avoid design rework during final stages

Method 03

Usability testing

Tested the final prototype with the same oncologists across the same 3 scenarios, measuring SUS and time-to-key-insight.

OUTPUT

SUS 52 → 80 (+28)

Turning insights into impact

Three observed frictions, three design responses, measured.

INSIGHT #1

Information was chronological, but labs and scans sat apart from the notes and diagnoses they belonged to

Each modality lived in a separate view. Physicians held the chronology in their head, switching tabs to verify whether a lesion grew between two scans.

WHAT I DESIGNED

Patient diagnosis above the fold: cancer type, size, location, and stage.

Live notes panel: AI drafts, doctor edits, every line links to its source.

Unified timeline: labs, scans, doses, decisions on one chronological view.

IMPACT

50%

faster review

60%

faster to key insight

"I can finally see how the tumor changed over the year on one screen."

INSIGHT #1

Information was chronological, but labs and scans sat apart from the notes and diagnoses they belonged to

Each modality lived in a separate view. Physicians held the chronology in their head, switching tabs to verify whether a lesion grew between two scans.

WHAT I DESIGNED

Patient diagnosis above the fold: cancer type, size, location, and stage.

Live notes panel: AI drafts, doctor edits, every line links to its source.

Unified timeline: labs, scans, doses, decisions on one chronological view.

IMPACT

50%

faster review

60%

faster to key insight

"I can finally see how the tumor changed over the year on one screen."

INSIGHT #2

Acting on a decision meant leaving the note

Ordering a scan or bloodwork meant exiting the chart and rebuilding context elsewhere. Drug safety got checked from memory, or downstream once the pharmacist caught it.

WHAT I DESIGNED

Inline ordering: tests and treatments placed without leaving the note, context pre-filled.

AI interaction check: The system flags any clashes with the patient's history along with reason for the doctor to verify

IMPACT

6-step task → 1 click

"I used to write a sticky note for the bloodwork. Now I click the order, and it tells me if there's a problem."

INSIGHT #2

Acting on a decision meant leaving the note

Ordering a scan or bloodwork meant exiting the chart and rebuilding context elsewhere. Drug safety got checked from memory, or downstream once the pharmacist caught it.

WHAT I DESIGNED

Inline ordering: tests and treatments placed without leaving the note, context pre-filled.

AI interaction check: The system flags any clashes with the patient's history along with reason for the doctor to verify

IMPACT

6-step task → 1 click

"I used to write a sticky note for the bloodwork. Now I click the order, and it tells me if there's a problem."

INSIGHT #3

Comparing scans across visits was the most critical step, and the slowest

To see whether a lesion grew, shrank, or appeared, oncologists opened two PACS windows and compared slice by slice. The decision moment of the visit was the most manual.

WHAT I DESIGNED

Comparing scans side-by-side with synced scroll that automatically align.

AI highlights changes between visits and oncologists can confirm it against the image.

IMPACT

4+ windows → 1 modal

"As oncologists we often refer to Xray specialists and this is a great way to get a quick insight. Will be useful during consultations to show the patients!"

INSIGHT #3

Comparing scans across visits was the most critical step, and the slowest

To see whether a lesion grew, shrank, or appeared, oncologists opened two PACS windows and compared slice by slice. The decision moment of the visit was the most manual.

WHAT I DESIGNED

Comparing scans side-by-side with synced scroll that automatically align.

AI highlights changes between visits and oncologists can confirm it against the image.

IMPACT

4+ windows → 1 modal

"As oncologists we often refer to Xray specialists and this is a great way to get a quick insight. Will be useful during consultations to show the patients!"

Reflection

"Fewer clicks" is the wrong goal if you strip context physicians unconsciously rely on. The timeline worked because it preserved chronology. The Smart Note worked because every AI claim ties back to its source, and because the next action lived where the information lived.

Next case study

Siebert Trading Platform →