A contact centre with no data centre

Cloud-native contact centre for a US healthcare provider, architected from an empty AWS account: routing, self-service, agent desktop, and the boundary that keeps patient data out of the parts that should never see it.

01/Context

The provider ran patient calls through a contact centre it could not change quickly. Adding a queue, a prompt, or a routing rule meant a vendor ticket and a release window. The brief was to replace it with something the provider's own teams could operate — and to do it without standing up a single server.

02/Constraints

Patient data: Anything that touches protected health information has to stay inside a small, auditable set of services. Most of the system is designed so that it never sees it at all.
No servers: The operating model assumed no infrastructure team. Every component had to be managed by AWS or it did not go in.
Business-hours change: Routing and prompts change with staffing. Those changes could not require a deployment.

03/Architecture

Telephony and routing: Amazon Connect owns the call. Contact flows stay deliberately thin — they identify the caller, ask Lambda what to do, and route. The decision logic lives in code that can be tested, not in a flow diagram that cannot.
Self-service: Amazon Lex handles intent capture for the calls that never need a person. Fulfilment runs in Lambda against the provider's systems of record, with a hard timeout and a human fallback on every path.
Services: .NET Core services behind API Gateway carry the domain: eligibility, appointments, case history. They are the only components inside the PHI boundary, and they are the only ones with access to the encrypted store.
Agent desktop: A React application embedding the Connect CCP, so an agent sees the caller's context in the same window as the call controls instead of alt-tabbing between four systems.

04/Outcome

A routing or prompt change went from a vendor ticket and a release window to something the provider's own operations team does itself, unattended.
No servers to patch: the whole system is managed AWS services plus deployable code.
The PHI boundary is small enough to review in an afternoon.

All case studies Next — Campaigns that decide in milliseconds