Challenge
Northwind Health, a multi-state primary-care network, had grown by acquisition. Each acquired clinic brought its own EHR, its own scheduling system, and its own labs integration. By the time the project started, clinicians were navigating between fourteen systems to complete a single patient encounter — five for the chart, three for orders, two for billing, four for ancillary results.
The cost of this fragmentation was not abstract. An internal time study found that primary-care physicians at Northwind were spending 47% of their patient-facing day on documentation and system navigation. Burnout scores were rising. Two well-respected physicians had left in the previous quarter, citing the technology environment specifically.
The previous integration vendor had spent eighteen months and $4.8M on a "unified portal" that clinicians described, in the post-mortem, as "the fifteenth system."
Approach
We approached this as a product problem first and a data problem second. Before any backend work, we ran two weeks of clinician shadowing across three sites and produced a single clinical workflow map: what a primary-care visit actually looks like, end to end, from the clinician's perspective.
That map drove every architectural decision. Data was unified only where the clinical workflow demanded it. Systems that served administrative needs — billing, credentialing — were left as separate logins. The unified surface was deliberately narrow: chart, orders, results, schedule. Everything else stayed where it was.
The platform itself: a FHIR-normalized data layer in AWS HealthLake, a Postgres operational store for the application, Snowflake for analytics. The clinician-facing application is a Next.js app with React Server Components, designed to feel like a single tool with one unified search and one unified navigation. HIPAA compliance was built in from the architecture phase — every PHI access is logged, every cross-system join is auditable, and the entire data layer runs in a dedicated AWS account with no inbound network access from the internet.
We staged the rollout deliberately. The first three months supported only chart-viewing — read-only, low-risk, fast feedback. Order entry shipped at month six. Billing integration followed at month nine. By the time the full workflow was live, clinicians had been using the platform for half a year and trusted it.
Outcome
The headline number, validated against the original time study: charting time per patient encounter dropped 62%. Clinicians now spend an average of 19% of their patient-facing day on documentation, down from 47%.
Northwind's clinician retention has stabilized for the first time in three years. The two physicians who had announced departures during the previous administration both rescinded their resignations during the rollout — they are still on staff. The CFO has signed off on a second-phase engagement to extend the platform to specialty clinics.
The system has been in production for nine months. Northwind's internal engineering team owns it; we operate as a small retainer for HealthLake schema evolution and FHIR mapping updates. The clinician adoption rate, which we feared would be the hardest part, settled at 94% within ninety days.
The work that mattered most, in retrospect, was not the data engineering. It was the discipline of saying no to integrating things clinicians did not want integrated. The platform succeeded because it was small.