Designing COMET at Amazon Care

The Breaking Point

Amazon Care was a hybrid telehealth and in-person primary care service—offering 24/7 virtual visits and mobile clinicians who could be dispatched to patients' homes across multiple states. As the service scaled rapidly from 100,000 to a projected 1 million members, the clinical operations infrastructure couldn't keep up.

John is staring at five spreadsheets, trying to confirm whether Dr. Martinez’s Washington license is about to expire. One system says March. Another says March 15. A third says March 30. Which one is right? This month alone, 94 licenses are expiring. They can’t risk scheduling a clinician with an expired license. That’s a compliance and patient safety issue. But they also can’t leave shifts unfilled as Amazon Care scales across 50 states. So they do what they always do: email three teams, wait, and hope someone has the answer.

This wasn't an edge case. This was every day. And it was about to get much worse. Amazon Care needed to onboard 1,000+ clinicians. Each new clinician required 12 hours of IT and informatics provisioning time. Each clinician needed 15 state licenses within 90 days of hire. Operations teams were drowning. This wasn't just operational inefficiency. It was a patient safety issue that was impacting real people—both the operations staff drowning in manual work and the clinicians trying to deliver patient care.

Making the Invisible Visible

Nobody knew the full scope. Eight teams each had their own piece of the puzzle. No one had put the pieces together.

I advocated for something that had never been done: a cross-functional workshop to map the entire clinician onboarding and licensing process end-to-end.

I brought representatives from all 8+ stakeholder teams into one room. These teams had never sat together like this. They each had their own systems, processes, language. Over the course of the workshop, we documented every touchpoint, every handoff, every dependency and pain point.

The result: The first comprehensive process map showing all tasks and team interactions across the entire clinician lifecycle.

The process map revealed invisible chaos that no single team could see on their own.

What the Map Revealed

Five critical problems emerged—problems that became our product roadmap:

1. Data duplication across 3-5 systems Clinician profiles existed in multiple systems with mismatched information. Name, credentials, specialties—manually entered into different tools. This wasn't just inefficient; it was a patient safety issue when information didn't match.

2. Manual spreadsheet tracking creating compliance risks Spreadsheets for medical license tracking had hit system limits twice. With 94 licenses expiring every month across 150 clinicians, there was no automated way to track expirations or send renewal reminders. The risk of clinicians treating patients with expired licenses increased with every new hire. This wasn't just a compliance issue—it was a patient safety issue.

3. Unsustainable onboarding burden Each clinician required 66 individual tasks taking 5 hours to 9 days to complete. 12 hours of IT and informatics provisioning time alone. Clinicians spent 90% of their time on credentialing and 50% on licensure applications—time taken away from patient care. To onboard 1,000+ clinicians, this manual process would require thousands of hours of operational overhead.

4. No single source of truth When different systems showed different license expiration dates, no one knew which was correct. Decision paralysis and compliance anxiety across operations teams.

5. Zero clinician visibility Clinicians had no way to see their own information, including where they were licensed to provide care. This lack of visibility put their licenses at risk if the system made mistakes and drove clinician dissatisfaction.

The Battle: Building the Right Platform

Armed with the process map, I had to fight for the right solution.

The Pressure to Integrate

Developers wanted to build into the existing clinical platform. One codebase, easier to maintain. But I knew this was wrong.

The clinical platform was built for patient-facing workflows—telehealth visits, scheduling appointments, viewing medical records. Consumer experiences.

Operations teams needed something fundamentally different: complex data tables showing hundreds of clinicians at once, bulk actions, role-based permissions, data-heavy interfaces for managing compliance across 50 states and tracking 94 license expirations monthly.

The Objections I Faced

Engineering leadership: "Building a separate platform means duplicate infrastructure, separate deployments, more maintenance overhead. Why can't we just add admin views to the existing system?"

Product leadership: "We're already stretched thin. Adding another platform to maintain will slow us down."

How I Won the Argument

I used the process map as my weapon. Eight teams. 66 onboarding tasks per clinician. 12 hours of IT provisioning. 94 licenses expiring monthly. 1,000+ clinicians to onboard. Then I showed them the clinical platform's design system—built for simple consumer flows.

The clincher was the design system research and the strategic initiative document. The business case classified this as a "large multi-year project" requiring a "unified clinician onboarding platform." An established admin-focused design system already existed, purpose-built for what we needed: complex data tables, bulk operations, role-based permissions, scalable information architecture.

My argument: "We're not building a separate platform because we want to. We're building it because operations teams need fundamentally different tools. The strategic initiative document calls for a unified platform approach. By using an established design system optimized for admin workflows, we'll move faster than trying to retrofit the clinical platform."

I designed with future workstreams in mind. The product vision for COMET (Care Operations Management Tool) explicitly positioned it as a "foundational platform for clinician-facing functions and automation"—validating the platform approach I advocated for.

The Outcome

I secured buy-in for COMET—a separate operations platform built on an admin-optimized design system. The platform thinking was validated almost immediately: when the scheduling team needed tools to plan clinician schedules, they built within COMET—proving the infrastructure could scale beyond the initial use case, exactly as the product vision intended.

The Solution: COMET

COMET (Care Operations Management Tool) became the single source of truth for all clinician data and operations.

Core capabilities:

Consolidated clinician profiles All clinician data in one system. No more hunting across five spreadsheets. One source of truth for 2,388 clinicians across 50 states.

Automated license tracking and compliance Automated detection of license expirations with renewal reminders. Automatic removal of clinicians from call queues if licenses expired, eliminating the patient safety risk. This replaced manual spreadsheet tracking of 94 monthly expirations.

Self-serve care team management Clinicians could update their own information, eliminating data duplication and reducing the burden on operations teams. Clinicians got visibility into their own profiles for the first time.

Streamlined onboarding Centralized data collection reduced the 66 individual tasks and 12 hours of IT provisioning time per clinician. Information entered once flowed to all systems, dramatically reducing the 5-9 day onboarding timeline.

Role-based permissions Different users could see and edit only what was relevant—practice managers saw their regions, compliance officers saw licensing data, clinicians saw their own profiles.

Standardized offboarding Proper access revocation for departing clinicians, eliminating security risks from ad hoc processes.

The admin-focused design system gave us the components to build quickly while maintaining consistency across data entry forms, information architecture, and complex data tables.

Impact

Amazon Care shut down in late 2022 as Amazon shifted focus to broader health initiatives like Amazon One Medical. While COMET didn't reach full production, the strategic work proved its value.

Patient safety and compliance: COMET automated license tracking, eliminating the risk of expired licenses. Automated reminders and call queue removal protected patients and clinicians.

Operational efficiency: Reduced IT provisioning time from 12 hours per clinician through automated workflows. Streamlined the 66-task onboarding process, enabling Amazon Care to scale from 100K to 1M members.

Clinician satisfaction: Automated 90% of credentialing work, giving clinicians their time back to focus on patient care instead of paperwork. Provided clinicians with visibility into their own profiles and licensing status for the first time.

Platform thinking validated: When the scheduling team needed tools, they built within COMET rather than creating their own system—proving the "foundational platform" approach could support multiple operational needs, exactly as the product vision intended.

Design system choice accelerated development: Adopting an established design system optimized for admin tools meant we could focus on workflows and logic rather than building components from scratch.

Process map became organizational artifact: The comprehensive process map became the shared language that aligned stakeholders and drove decision-making across all 8+ teams. The product vision document later confirmed what the workshop revealed—66 individual tasks, spreadsheets hitting limits twice, and compliance risks from lack of single source of truth.

What I Learned