You'll need a password to unlock this case study!

Flowsheets

Flowsheets

From zero definition to $5.6M in pipeline →
How I took medical flowsheet extraction from raw capability to shipped product in under three months.

From zero definition to $5.6M in pipeline →
How I took medical flowsheet extraction from raw capability to shipped product in under three months.

ROLE

Lead Product Designer

TIMELINE

< 3 months

COMPANY

Supio

SHIPPED

Jan '26

Summary

01 Problem

02 Research

03 Design

04 Collaboration

05 Result

Medical malpractice attorneys were asking for a tool Supio didn't have. Half of the story of a malpractice case lives inside flowsheets — dense nursing records that track vitals, medications, and treatments minute-by-minute — but those documents arrived as thousands of cells locked in PDFs.

As the sole designer, I took the feature from zero defintion to a shipped product in under three months: conducting user research, defining the information architecture, making key calls on AI trust and workflow design, and coordinating across a team in three locations.

Medical malpractice attorneys were asking for a tool Supio didn't have. Half of the story of a malpractice case lives inside flowsheets — dense nursing records that track vitals, medications, and treatments minute-by-minute — but those documents arrived as thousands of cells locked in PDFs.

As the sole designer, I took the feature from zero defintion to a shipped product in under three months: conducting user research, defining the information architecture, making key calls on AI trust and workflow design, and coordinating across a team in three locations.

THE PROBLEM

THE PROBLEM

Two avenues of truth, but only one was accessible.

Two avenues of truth, but only one was accessible.

Building a malpractice case requires reconciling two different records: medical records, the hospital's documented account of events, and flowsheet measurements, the objective clinical data that doesn't lie.

Before this feature, attorneys had the medical records, and Supio's AI medical chronology, which extracted and displayed individual medical events as distinct data structures in chronological order. The flowsheets, in contrast, were dense, often spanned hundreds of pages, and were a pain to read.

The firms asking for a solution represented over $5.6M in potential contracts.

Building a malpractice case requires reconciling two different records: medical records, the hospital's documented account of events, and flowsheet measurements, the objective clinical data that doesn't lie.

Before this feature, attorneys had the medical records, and Supio's AI medical chronology, which extracted and displayed individual medical events as distinct data structures in chronological order. The flowsheets, in contrast, were dense, often spanned hundreds of pages, and were a pain to read.

The firms asking for a solution represented over $5.6M in potential contracts.

  • To prove malpractice, attorneys must show:

    1. Standard of care

    What should have happened - the accepted medical protocol

    1. The Breach

    What actually happened, and where care deviated.

    1. Timeline of knowledge

    Exactly when the provider knew or should have known

    1. Causation

    How the breach directly resulted in patient harm.

  • Where malpractice hides

    WHAT HAPPENED

    WHAT IT LOOKS LIKE IN THE DATA

    Protocol not followed

    Events documented, no flowsheets measurements to back them up

    Missed signs of serious condition

    Abnormal measurements, no event acknowledgement to match

    Medication prescribed wrong

    Abnormal measurements, no obvious event attached

    Critical lab results not reviewed

    Abnormal lab values with no events documenting physician review

THE RESEARCH

We had the extraction capability. We had zero idea what to do with it.

We had the extraction capability. We had zero idea what to do with it.

I was brought on as the sole designer with no prior work to build on. The engineering team could extract a flowsheet. No workflow, no IA, no definition of what attorneys would do with the data once they had it.

I was brought on as the sole designer with no prior work to build on. The engineering team could extract a flowsheet. No workflow, no IA, no definition of what attorneys would do with the data once they had it.

My first job was to go figure it out.

My first job was to go figure it out.

I started with CS and sales — closest to the customer requests — then moved to user interviews and sales call synthesis. The defining insight:

I started with CS and sales — closest to the customer requests — then moved to user interviews and sales call synthesis. The defining insight:

There is no single way a malpractice attorney works.

There is no single way a malpractice attorney works.

2

Medical malpractice attorneys interviewed

12

Sales calls with medical malpractice firms analyzed

Hypothesis first

Started with suspected error — missed medication, abnormal vital — and work backward through the data.

Big picture first

Pull the full extraction, scan for patterns, then narrow down to what matters for the case narrative.

Expert handoff

Need clean, exportable data for medical experts. Just need the tool to get out of their way.

This killed the idea of a guided, deterministic flow. We could design a wizard for "the" way to analyze a flowsheet. That workflow doesn't exist, and ascribing one might create more distrust than value.


The design had to open, giving attorneys full control of their data and the flexibility to approach it however their case demanded.

THE DESIGN

Design Process

Ideation & alignment

Generated a PRD using meeting notes and transcripts in ChatGPT. This helped ensure alignment on the core functionalities we needed.

Wireframes, prototypes in Claude Code

Used Claude to generate possible approaches to the IA: how can we bring together an existing medical chronology and a dense table extraction?


I moved the prototype to Claude Code for refinements, and consolidated all approaches into a clickable prototype I could bring to stakeholder meetings.


Refining in Figma

  • Finalized the separation of the Flowsheets and existing medical chronology surface.

  • Introduced Manage Columns, View Source, Insert & Edit Cells, Filtering, and Export as the data management action set

  • Spec'd out Highlight Cells, Pin-to-timeline, and a joint "Pin highlighted cells" function that would allow users to find significant cells and add them to the case narrative.

Prototyping in Claude Code

Exploring navigational elements using Claude Code

Bringing together multiple options side-by-side

THE DESIGN

Quickly iterating on prototypes in Claude gave us the clarity we needed.

Quickly iterating on prototypes in Claude gave us the clarity we needed.

The primary requirement for this feature was that the user had to be able to view the flowsheet data in chronological order with the timeline events. Stakeholders were hesitant to touch our existing medical chronology, but quick Claude prototypes showed that we would just be building the same features all over again.

The primary requirement for this feature was that the user had to be able to view the flowsheet data in chronological order with the timeline events. Stakeholders were hesitant to touch our existing medical chronology, but quick Claude prototypes showed that we would just be building the same features all over again.

Engineering constraint
Adding too much data to the timeline carried too much dependencies.

From a user perspective:
Too much flowsheet data would also generate noise.

Pin-to-timeline was scoped as a deliberate, user-initiated action rather than automatic population.

Engineering constraint
Adding too much data to the timeline carried too much dependencies.

From a user perspective:
Too much flowsheet data would also generate noise.

Pin-to-timeline was scoped as a deliberate, user-initiated action rather than automatic population.

MVP DECISION

Extend the existing timeline

Extend the existing timeline

Attorneys already needed most of their chronology in the narrative — that's why our medical chronology existed, and why it was so successful in the market. Pin-to-timeline connected individual flowsheet measurements to the surface attorneys already knew, with minimal timeline changes.

Attorneys already needed most of their chronology in the narrative — that's why our medical chronology existed, and why it was so successful in the market. Pin-to-timeline connected individual flowsheet measurements to the surface attorneys already knew, with minimal timeline changes.

Pros:
✓  Leaner MVP

✓  No redundancy

✓  Attorneys already knew the surface

Pros:
✓  Leaner MVP

✓  No redundancy

✓  Attorneys already knew the surface

Cons:
✕  Extra click to navigate between two tabs

ORIGINAL HYPOTHESIS

New dedicated surface

New dedicated surface

Side-by-side view: scroll flowsheet and timeline simultaneously, pick what to include in the case narrative from each pane.

Side-by-side view: scroll flowsheet and timeline simultaneously, pick what to include in the case narrative from each pane.

Pros:
✓  Reduces clicks between the two artifacts

✓ View flowsheet and medical events on one timeline

Pros:
✓  Reduces clicks between the two artifacts

✓ View flowsheet and medical events on one timeline

Cons:
✕  Redundant with existing timeline

✕  Added manual work for attorneys

✕  Two surfaces to build and maintain

Cons:
✕  Redundant with existing timeline

✕  Added manual work for attorneys

✕  Two surfaces to build and maintain

THE DESIGN

THE DESIGN

Feature Design

Feature Design

01

01

Turn the PDF into something attorneys can touch

Turn the PDF into something attorneys can touch

AI-powered extraction converted dense PDFs into an interactive, editable table — dates and times anchored in the first column, metrics across the top. A document that previously required hours of manual transcription became searchable, sortable, and manipulable in seconds.

Cells were editable by design. AI output should never be ground truth in a legal context — one wrong value could undermine a case.

AI-powered extraction converted dense PDFs into an interactive, editable table — dates and times anchored in the first column, metrics across the top. A document that previously required hours of manual transcription became searchable, sortable, and manipulable in seconds.

Cells were editable by design. AI output should never be ground truth in a legal context — one wrong value could undermine a case.

02

02

Help them find the abnormalities — not just the data

Help them find the abnormalities — not just the data

General flowsheet data isn't useful on its own. Normal measurements that match the timeline don't advance the case — they create noise. The design goal was to surface abnormalities: the measurements that don't match, the gaps between what was recorded and what was acknowledged.

Four interconnected capabilities, each designed to support every type of attorney workflow.

General flowsheet data isn't useful on its own. Normal measurements that match the timeline don't advance the case — they create noise. The design goal was to surface abnormalities: the measurements that don't match, the gaps between what was recorded and what was acknowledged.

Four interconnected capabilities, each designed to support every type of attorney workflow.

Base flowsheet functions

Flowsheet investigation user flow

Flowsheet investigation user flow

THE DESIGN

Features

FEATURE 01

FEATURE 01

Conditional Highlighting

Conditional Highlighting

Define quantitative rules ("fetal heart rate < 110") or qualitative rules ("is empty") to surface abnormal values across thousands of cells instantly. Flagged cells get distinct visual treatment.

Define quantitative rules ("fetal heart rate < 110") or qualitative rules ("is empty") to surface abnormal values across thousands of cells instantly. Flagged cells get distinct visual treatment.

FEATURE 02

FEATURE 02

Pin to Timeline

Pin to Timeline

Pull critical measurements directly into the medical chronology as standalone events. Measurements appear distinctly from documented events, letting attorneys see what was measured against what was acknowledged, temporally.

Pull critical measurements directly into the medical chronology as standalone events. Measurements appear distinctly from documented events, letting attorneys see what was measured against what was acknowledged, temporally.

FEATURE 03

FEATURE 03

AI Chat

AI Chat

Ask natural language questions (e.g. "Are there any abnormal values?") across flowsheets, files, and chronologies simultaneously. Surfaces leads for where to start looking, and what might have been missed.

Ask natural language questions (e.g. "Are there any abnormal values?") across flowsheets, files, and chronologies simultaneously. Surfaces leads for where to start looking, and what might have been missed.

FEATURE 04

FEATURE 04

CSV Export

CSV Export

Hand off clean, structured data to medical experts without leaving the platform. Critical for the expert-handoff workflow, and a net-new capability that expended platform value beyond malpractice.

Hand off clean, structured data to medical experts without leaving the platform. Critical for the expert-handoff workflow, and a net-new capability that expended platform value beyond malpractice.

Conditional Highlighting

Conditional Highlighting

Pin-to-timeline

Pin-to-timeline

AI Chat

AI Chat

Figma Refinements + Handoff

Figma Refinements + Handoff

"This is a game changer. Holy S**T — the exact data I need to disprove opposing counsel."

"This is a game changer. Holy S**T — the exact data I need to disprove opposing counsel."

A medical malpractice attorney, reacting live during a feature demo. Requested platform access before the end of the demo.

A medical malpractice attorney, reacting live during a feature demo. Requested platform access before the end of the demo.

THE COLLABORATION

Three time zones, one tight deadline

Three time zones, one tight deadline

This project was the most demanding cross-functional coordination I've navigated. Mid-project, the engineering team shifted; at one point we were building across three time zones with part of the team in China.

I communicated directly in Mandarin with one engineer when that was the clearest path forward. The lesson: Collaboration isn't about keeping people informed. It's about making sure everyone understands the why.

When the reasoning is clear, handoffs are smooth.

Functions coordinated

Functions coordinated

  • Engineering

    Design intent across 3 time zones, including direct Mandarin communication


  • Product

    Scope definition, MVP trade-offs, engineering constraint navigation


  • CS & Sales

    Customer pain points, demo alignment, go-to-market prep


  • Product Marketing

    Positioning as first-to-market differentiator in med-mal

  • Engineering

    Design intent across 3 time zones, including direct Mandarin communication


  • Product

    Scope definition, MVP trade-offs, engineering constraint navigation


  • CS & Sales

    Customer pain points, demo alignment, go-to-market prep


  • Product Marketing

    Positioning as first-to-market differentiator in med-mal

THE RESULT

THE RESULT

Impact

Impact

$5.6M

Pipeline unlocked at launch

First

Platform with flowsheet extraction for legal teams

3

New markets: med-mal, birth injury, nursing care

Jan '26

Shipped as private beta in under 3 months.

$5.6M

Pipeline unlocked at launch

3

New markets: med-mal, birth injury, nursing care

First

Platform with flowsheet extraction for legal teams

Jan '26

Shipped as private beta in under 3 months.

$5.6M

Pipeline unlocked at launch

3

New markets: med-mal, birth injury, nursing care

First

Platform with flowsheet extraction for legal teams

Jan '26

Shipped as private beta in under 3 months.

What went well

What went well

  • Ffeature became Supio's primary differentiator in the medical malpractice, birth injury, and nursing care markets.


  • Private beta framing set honest expectations and gave us real extraction data to design the error-handling layer from


  • Editable cells proved more critical than anticipated as extraction accuracy varied across healthcare systems.

  • Ffeature became Supio's primary differentiator in the medical malpractice, birth injury, and nursing care markets.


  • Private beta framing set honest expectations and gave us real extraction data to design the error-handling layer from


  • Editable cells proved more critical than anticipated as extraction accuracy varied across healthcare systems.

What I'd do differently

What I'd do differently

We entered with a narrow definition because we were designing for one specific firm. That focus helped us move fast — but flowsheet formats vary significantly across healthcare systems and case types. A broader research sample early on, even a handful of interviews across different firm types, would have given us a more durable foundation. In specialized domains, the cost of a narrow sample compounds as you scale.

We also realized, much later, that our feature had many more extraction capabilities than just flowsheets. This feature now lives as "Tabular Analysis," a tool that can help extract any sort of data in a table format. Making this pivot earlier could've helped us establish better expectations around different flowsheet formats, while helping define other possible use cases.

We entered with a narrow definition because we were designing for one specific firm. That focus helped us move fast — but flowsheet formats vary significantly across healthcare systems and case types. A broader research sample early on, even a handful of interviews across different firm types, would have given us a more durable foundation. In specialized domains, the cost of a narrow sample compounds as you scale.

We also realized, much later, that our feature had many more extraction capabilities than just flowsheets. This feature now lives as "Tabular Analysis," a tool that can help extract any sort of data in a table format. Making this pivot earlier could've helped us establish better expectations around different flowsheet formats, while helping define other possible use cases.

Hi! I'm Elisa. Let's build something together.

Message me