Its own leaders rated the platform 4 out of 10.
Medicare enrollment season was coming anyway.

A routing change can break the agent desktop or a downstream integration, so the assessment couldn't look at contact flows alone. We traced contact flows, queues, Lambda functions, CRM dependencies and the agent desktop together, reviewed against AWS best practices and DORA benchmarks:

• Engineers reviewed failover readiness, deployment practices and protected-data handling, then sequenced remediation by impact and effort.
• Program configuration data differed across health-plan clients, so we mapped those differences before restructuring the data used for reporting.
• A release audit identified the missing PR workflows, CI/CD steps and audit trails needed to trace a change from review to deployment.
• The compliance posture was mapped against HIPAA, SOC 2, ISO 27001/27002 and HITRUST R2.
• The desktop replacement roadmap, from the third-party agent desktop to Amazon Connect's own, recorded dependencies and the work needed to keep each client separate during migration.

The four weeks produced a current-state report, a risk-mitigation plan, an architecture diagram, a configuration workbook and an execution plan, and confirmed the team structure on evidence.

Many of Amazon Connect's capabilities had simply never been switched on. The retained Connect pod improved inbound, outbound, task and guided contact flows. Agent Workspace Views with dynamic data bindings and HTML sections bring each client's information onto the desktop during enrollment calls, so people type less mid-call.

We connected the operation to the platform's supervision tools:

• Contact Lens supplies in-call and post-call analysis to check the required Medicare disclaimers, as the call happens rather than after.
• CloudWatch dashboards show queue depth, handling times, people's performance and escalation patterns while the operation runs.
• Restructured program configuration gives consistent reporting across clients, so supervisors can compare activity without first reconciling different data definitions.

The fixes only stay fixed if the next change can't undo them. Every change now goes through PR review, version control and CI/CD, with an audit trail. Development and testing are separated from production access: offshore engineers develop and test, and designated onshore engineers perform production deployments under the customer's protected-data controls, so PHI never reaches the offshore environment.

Why: in a regulated contact center, how a change gets released is something the compliance team has to be able to show.

A call can end without the business knowing whether the work is done. An unanswered attempt may need a retry, a suppression rule may make the next attempt ineligible, and an incomplete disposition may need reconciling. The previous design used Amazon Connect as both the phone system and the campaign manager, which is where the fragility came from.

We gave every outbound item a persistent lifecycle outside the individual contact flow. Amazon Connect handles dialing, routing and agent availability. The orchestration layer decides which work is eligible and what happens after each attempt:

• Campaign records enter through S3, Lambda and DynamoDB, with schema validation, phone normalization, client and program identification, suppression checks and idempotent deduplication before a record becomes eligible work.
• A state machine and policy matrix represent attempt limits, cooldowns, do-not-call rules, time zones, holidays, state-level dialing rules and contact-fatigue limits, applied to the work item across all its attempts.
• The scheduler picks eligible items by client and program, and a lease-and-lock mechanism coordinates concurrent campaign processes, cutting duplicate-dial risk.
• Attempt metadata drives the next state, and a reconciliation job handles incomplete or missing outcomes, so the record no longer depends on manually entered dispositions alone.
• Reporting follows items through ingested, ready, attempted, completed and suppressed states.

Because the rules live in a state machine, they hold on the busiest day whoever is on shift.