eConsult
AI-assisted triage for faster, safer online consultations
Year
2024
Industry
Health-Tech
Role
Design lead
Overview
At eConsult, clinicians were spending too much time manually reviewing and assigning online consultations before care could begin.
I designed an AI-assisted triage workflow that suggested relevant outcomes, reduced repetitive effort and helped clinicians move faster — while keeping every clinical decision under their control.
Team
Product Manager · ML Engineers · Software Engineers · Clinical stakeholders
Focus
AI-assisted workflows · Healthcare · Product design
Impact
46%
faster triage
40+
interactions reduced to 5–20
~22
hours saved weekly per practice
Challenge
Online consultations were helping patients reach their GP practices more easily, but they created a growing operational challenge for clinicians.
Each request still needed to be reviewed, assessed for urgency, assigned to the right team and documented manually. The process was repetitive and time-consuming, but also clinically sensitive.
Introducing AI could reduce effort, but only if clinicians remained able to understand, verify and override every recommendation.
How might we make triage faster without compromising clinical judgement or patient safety?
Research
I worked with the product team to understand how clinicians currently reviewed online consultations and what would make AI support feel safe enough to use.
Research showed that clinicians were open to support with repetitive triage work, but they did not want a system making opaque decisions on their behalf. Their trust depended on being able to verify recommendations against the patient’s original information.
This gave us a clear direction:
The AI should reduce effort, not remove judgement.
Research Approach
I partnered with the Product Manager to conduct onsite and remote research with clinicians across approximately 20 practices.
We also analysed more than 50 existing triage sessions to understand where time and interaction effort were being lost.
The research focused on:
how clinicians completed triage today
which steps created the most friction
what information they needed to make decisions confidently
what would make AI recommendations feel useful or unsafe
Key research findings
Three findings shaped the direction of the product:
Finding 1: Clinicians needed to identify relevant signals faster
Consultations could contain detailed or messy patient information, increasing the effort required to assess the right outcome.
Finding 2: Repetitive decision assembly created avoidable work
Clinicians were manually completing multiple fields even when a likely triage outcome was already clear.
Finding 3: Verification was essential for trust
Clinicians wanted access to the original patient input behind any recommendation rather than relying solely on an AI-generated explanation.
Solution
I designed a workflow where AI suggested up to two complete triage options based on the patient’s consultation.
Each recommendation included the appropriate urgency, acuity, consultation mode and team assignment. When a clinician selected an option, the relevant fields were automatically populated for review.
Rather than assembling each triage decision from scratch, clinicians could start from a useful recommendation, adjust anything necessary and save.
Most importantly, the system was designed as decision support, not automated decision-making.
Key design decisions
I partnered with the Product Manager to conduct onsite and remote research with clinicians across approximately 20 practices.
We also analysed more than 50 existing triage sessions to understand where time and interaction effort were being lost.
The research focused on:
how clinicians completed triage today
which steps created the most friction
what information they needed to make decisions confidently
what would make AI recommendations feel useful or unsafe
Why only two recommendations?
I partnered with the Product Manager to conduct onsite and remote research with clinicians across approximately 20 practices.
We also analysed more than 50 existing triage sessions to understand where time and interaction effort were being lost.
The research focused on:
how clinicians completed triage today
which steps created the most friction
what information they needed to make decisions confidently
what would make AI recommendations feel useful or unsafe
Designing for clinical trust
The most important design challenge was not displaying an AI suggestion. It was helping clinicians verify why that suggestion had been made.
I designed a jump-to-source evidence pattern that allowed clinicians to inspect the original patient information connected to a recommendation.
This meant clinicians could validate suggestions against the source rather than relying on a simplified AI explanation alone. In a clinical workflow, transparency was not a secondary feature. It was central to safe adoption.
Testing and iteration
I partnered with the Product Manager to conduct onsite and remote research with clinicians across approximately 20 practices.
We also analysed more than 50 existing triage sessions to understand where time and interaction effort were being lost.
The research focused on:
how clinicians completed triage today
which steps created the most friction
what information they needed to make decisions confidently
what would make AI recommendations feel useful or unsafe
Impact
The workflow reduced unnecessary effort while preserving clinician control.
46.6% reduction in average triage time
Average triage time reduced from 5 minutes 38 seconds to just under 3 minutes.
40+ interactions reduced to 5–20
Clinicians needed significantly fewer steps to complete a typical triage decision.
~22 hours saved weekly per practice
Reduced triage effort created more capacity for patient-focused clinical work.
The result was not simply a faster workflow. It was an AI-supported clinical experience designed around responsible decision-making.
Measuring success
I partnered with the Product Manager to conduct onsite and remote research with clinicians across approximately 20 practices.
We also analysed more than 50 existing triage sessions to understand where time and interaction effort were being lost.
The research focused on:
how clinicians completed triage today
which steps created the most friction
what information they needed to make decisions confidently
what would make AI recommendations feel useful or unsafe
Reflection
This project reinforced that designing AI for high-stakes environments is primarily about designing for trust.
The hardest part was not creating recommendations. It was deciding how to introduce them without weakening professional judgement, how to make them verifiable and how to ensure efficiency never came at the cost of safety.
The experience taught me that AI becomes genuinely useful when it helps experts move faster while giving them every reason to remain confidently in control.