Cloud adoption in healthcare has accelerated faster than anyone predicted. Between the telehealth surge of 2020 and the staffing shortages that followed, hospital systems found themselves simultaneously overloaded and under-resourced — a combination that made the old model of human-staffed call centers untenable. The result has been a wave of cloud migrations for patient engagement workloads that would have seemed aggressive even two years ago.
We spoke with IT leaders at three health systems — a regional hospital network, a multi-site ambulatory practice, and a specialty telehealth provider — about what the shift actually looks like from the inside.
The Compliance Question Everyone Asks First
HIPAA doesn't prohibit cloud hosting of patient data — but it does require that any cloud vendor sign a Business Associate Agreement (BAA) and demonstrate appropriate technical safeguards. Every IT leader we spoke with said the compliance conversation was the first obstacle, and that it took longer than expected.
"We spent six weeks in legal before a single line of configuration was written," said the VP of Digital Health at the regional network. "But once we had the BAA signed and the data flow mapped, the actual technical work was straightforward."
Key safeguards they implemented before go-live:
- End-to-end encryption for all data in transit and at rest
- Role-based access controls mapped to existing Active Directory groups
- Audit logging for every patient data access event, retained for 7 years
- Automatic de-identification of transcripts before any third-party analytics processing
What Moved to the Cloud First
None of the organizations started with their most sensitive workloads. The typical migration sequence looked like this:
- Appointment reminders and confirmations — outbound only, low PHI exposure, high volume
- Post-discharge follow-up calls — structured scripts, measurable outcomes (readmission rates)
- Lab result notifications — triggered by EHR events, required careful HL7 integration
- Inbound appointment scheduling — highest complexity, tackled last
The ambulatory practice went live with appointment reminders in 90 days. It took another eight months to reach inbound scheduling. "The first phase proved the model to the board," their CTO told us. "Everything after that was execution."
The ROI Was Faster Than Expected
All three organizations reported positive ROI within the first year — faster than their pre-migration projections. The primary driver wasn't cost reduction (though that happened); it was capacity expansion. The regional network added the equivalent of 14 FTE call-center capacity without adding headcount, absorbing a post-merger patient volume spike that would have otherwise required an emergency hiring push.
"We didn't get rid of any staff. We redeployed them. The agents that used to answer 'what time is my appointment' now handle escalations, insurance questions, and care coordination. The work is harder, but people are more engaged."
— VP of Patient Experience, regional hospital network
What Surprised Them
Two surprises came up in every conversation. First: patient acceptance was higher than expected. Contrary to assumptions, older patient demographics were not the resistance point — the friction came from clinical staff who worried about edge cases the AI couldn't handle. Solving that required clear escalation paths and transparency about what the agent would and wouldn't attempt.
Second: the EHR integration was the longest pole in the tent. Epic, Cerner, and Athena each have different webhook patterns and rate limits, and the integration work consumed more engineering time than the AI configuration did. Teams that used a pre-built EHR connector (rather than rolling their own) cut that timeline by roughly half.
