Innovaccer x Ai Copilot • project 2025
AI Copilot for faster clinical decisions and closing care gaps
ROLE
Product Designer
TIMELINE
October 2024 - May
2025
TEAM
1 PM
2 Devs
1 Designer (me!)
SKILLS
Product Design
Product Strategy
Prototyping
Context
What is the Provider Copilot at Innovaccer?
An AI-powered, EHR-agnostic solution built to embed evidence-based clinical guidelines (like HIPAA guidlenes) directly into healthcare workflows.
Cross-Platform Clinical Guidelines
Partnered with Innovaccer Inc. and a leading U.S. health system to extend proprietary Value-Based Care Guidelines beyond specific Epic environments.
Why It Mattered
Inconsistent adherence to clinical guidelines drives uneven, inefficient care across American health systems.
The Opportunity
Turn extensive best-practice repositories into guidance clinicians can actually use at the point of care.
PROBLEM
Care guidelines were hard to see and harder to use in the moment of patient care.
This is the standard setup of any physician/provider in America during patient interaction.
Manual Effort
Providers had to dig through external libraries, map recommendations to patient context, then transcribe actions back into the EHR by hand.
System Fragmentation
Guidelines were often locked inside specific EHRs like Epic, out of reach for providers on other systems.
Cognitive Burden
Between high patient volume and dense documentation, clinicians struggled to act on recommendations in the moments that mattered most.
Research
Combined a literature review of clinical barriers with primary interviews.
Six semi-structured, one-on-one interviews (three clinicians and three clinical staff) to map typical visit workflows.
Primary findings were as followed:
No Time
65% of physicians pointed to lack of time as the main reason for skipping guidelines.
No Context
Existing systems couldn't connect clinical content to real-time patient context.
Broken Flow
Clinicians hit a "broken workflow" every time they had to leave the EHR to search externally.
SOLUTION
Provider’s perfect Copilot
An AI assistant embedded directly into the Pre-Visit Summary, working off real-time FHIR data.
CORE FLOWS
It Scans
Automatically interprets symptoms, vitals, and lab results using NLP and rule-based care gap detection.
It Surfaces
Matches patterns to the right specialty, then narrows down to the specific clinical condition using a hybrid guideline-matching engine, only surfacing recommendations above 90% confidence.
It Acts
Suggests Next Best Actions, labs, imaging, meds, validated against patient-specific factors like allergies and renal function. Providers add them to their cart, sign off, and orders route straight to the EHR via FHIR ServiceRequest.
OUTCOMES
Providers rated the guidelines as clinically relevant and easy to act on with a 78% Satisfaction lvl.
64% Faster
Time spent applying guidelines per visit dropped from 7 minutes to 2.5 minutes.
72% Adherence
Guideline adherence rose from 52% to 72% post-implementation.
45% Quicker Decisions
Providers moved through complex clinical cases significantly faster.
DESIGN DECISIONS
I identified the most intuitive way to bring this
Copilot to life.
With insights and iteration from prototyping and testing, we got to thinking how a clear, intuitive solution could be integrated within the existing provider’s digital ecosystem.
Insight 1: No New Windows
Providers won’t tolerate a separate tool.
Action
Used an Electron-based overlay that launches guidelines right alongside the EHR, keeping providers in one ecosystem.
Insight 2: (×_×;) transcription
Manual transcription was the biggest pain point.
Action
Prioritized a context-aware transcription widget.
Insight 3: Three clicks, Max….
Providers wanted minimal interaction.
Action
Insight 4: Fit in, don’t interrupt
Disruption had to stay low.
Action
Moved away from intrusive pop-ups and built guidelines directly into the PVS, a tool providers already use daily.
ITERATIONs
Entry Points
Started with universal access through settings, rejected it for patient-banner icons since settings lacked clinical context.
UI Surfaces
Moved from side widgets, easily ignored, to highlight cards, and finally to a collapsible PVS card that fit quietly into tertiary workflows.
Automation Levels
Evolved from a view-only manual search to a fully automated system that identifies specialties and conditions straight from unstructured data.
REFLECTION
What I learned
Speed and control pulled against each other (×_×;)
The fastest flow would let the AI place orders on its own, but….
every doctor I designed for needed to stay the one who decides. Making them add to a cart and sign off adds a few clicks, but it’s what lets them act on the AI’s suggestions at all.
saraSanwal2026™