140 servers out of a data center, 11 months

Challenge

The network ran 140 physical and virtual servers in a leased data center. The lease ended in 14 months and the landlord would not extend. A HIPAA audit was scheduled for month seven. The application inventory lived in three spreadsheets that disagreed with each other, and two clinical systems had vendors who had not been contacted in years.

The CTO needed one plan that satisfied the auditor, the clinicians and the finance team, and a partner who would still be there for the last wave.

Approach

Three weeks of discovery produced a single inventory of 31 applications, 18 databases and every dependency between them, built by scanning the network rather than asking people to remember. Each application got a disposition against the 7 Rs. Nine were retired outright. Four were repurchased as SaaS. The rest were rehosted or replatformed onto AWS.

We built the landing zone first: an AWS Organization with separate accounts for clinical, corporate and shared services, Terraform for everything, Okta for identity, and the HIPAA controls (encryption, logging, access reviews) as code before any workload landed. The auditor reviewed the landing zone in month six and signed off before the first clinical cutover.

Migration ran in six waves of two to four weeks. Every wave was rehearsed in a staging account the week before, with a timed rollback. Databases moved with AWS Database Migration Service in continuous replication so cutover was a DNS change and a final sync. The two clinical systems moved last, on a Saturday night, with the vendors on the bridge.

Result

Zero minutes of unplanned downtime across all six cutovers. The run-rate fell 34 percent after rightsizing and reserved capacity, against a 20 percent target. The lease was handed back a month early. The architecture decision record from month two is still the document the network uses to brief new auditors, and the managed services team that ran the cutovers now runs the estate under an Enterprise SLA.