A credit union migrates an ~800 GB core-banking database to AWS behind five gates

The customer is a member-owned U.S. credit union regulated under NCUA Part 748 and its Appendix A safeguards, the FFIEC IT Examination Handbook, and the GLBA Safeguards Rule, serving members through always-on digital and core-banking channels. The production estate behind those channels ran on aging on-premises infrastructure: a Microsoft SQL Server 2016 core database platform on a 16-core blade carrying a production database of roughly 800 GB, the Jack Henry core-banking system databases, a custom-application database, and SSIS (data integration) and SSRS (reporting) running on the same platform — with analytics users querying the production OLTP database directly, contending with the member-facing workload. An imminent SQL Server Enterprise license renewal and a looming hardware refresh made the on-premises status quo an expensive default.

Alongside the on-premises core, the credit union already ran a growing AWS estate — two production applications on Amazon ECS with AWS Fargate, a relational tier on Amazon RDS, scaled AWS Glue ETL, Amazon QuickSight, and a Terraform-deployed data lake — sitting in us-west-1 with multi-region transit-gateway complexity and three separate CloudTrail trails. That estate had grown without an independent, framework-driven assessment, and with only two people actively managing AWS the institution carried key-person dependency and limited capacity to formalize incident response, disaster recovery, and operational readiness. For a financial institution handling member financial data, undocumented gaps in IR, DR, and observability are compliance and member-trust exposure, not just technical debt.

VeUP ran a full AWS Well-Architected Framework Review across all six pillars as a structured, multi-session engagement using the AWS Well-Architected Tool question set, producing an executive findings report and a phased 90-day remediation roadmap sequenced into Immediate (0–30 days), Short-term (30–60 days), and Medium-term (60–90 days) workstreams. The review prioritized five HIGH-risk findings — no incident-response program, an absent disaster-recovery strategy, IAM deficiencies, deployment and operational-readiness gaps, and insufficient observability — each mapped to a specific Well-Architected question and a concrete AWS-native remediation. Off the back of the review, VeUP delivered a 3-tier Security Accelerator, with Tier 1 establishing the security foundation: an Incident Response Plan aligned to NIST SP 800-61, break-glass IAM roles, annual tabletop exercises, and Amazon GuardDuty plus AWS Security Hub enablement.

In parallel, VeUP scoped and delivered the migration of the credit union’s Jack Henry core-banking SQL Server database into AWS — a cost-comparison-driven move, executed as part of a wider phased migration of the on-premises estate. The migration program follows a single architecture of record — agreed with the customer in February 2026 and pressure-tested in the March Well-Architected Review: the core SQL Server platform replatforms from SQL Server 2016 Enterprise to the AWS license-included SQL Server 2022 Standard AMI on Amazon EC2, while the Jack Henry core-banking databases rehost to EC2 with minimal change and real-time connectivity preserved over an encrypted site-to-site VPN, coordinated with Jack Henry. Amazon RDS was evaluated for the core platform and deliberately not selected — SSIS and SSRS require the full, self-managed SQL Server feature set that a managed database service does not expose in the form this estate needs — a deliberate platform call that keeps the regulated core stable while still capturing the licensing, right-sizing, and governance benefits of the move.

Discovery and assessment ran as a structured readiness engagement built on the AWS Cloud Adoption Framework and the Well-Architected Review. All six CAF perspectives — Business, People, Governance, Platform, Security, and Operations — were scored current-versus-target and walked through with the customer: where the institution stood, where the gaps were, and what to do about them. Every workload got a 7Rs disposition; every risk, a quantified place on the heat map; every recovery objective, a tier. Controls were mapped first to NIST CSF, CIS, and the Well-Architected Security pillar, then to the rules a credit union answers to: NCUA Part 748 and Appendix A, FFIEC, and the GLBA Safeguards Rule. The readiness verdict was explicit: ready to migrate under a phased plan, once the flagged governance and security items closed before cutover.

Pre-cutover mobilization closes those items: a separate, locked-down production AWS account where deployment happens only through Terraform and the CI/CD pipeline — no console access — so every production change is code-reviewed, version-controlled, and auditable; privileged access vaulted and brokered through CyberArk with session recording and a documented break-glass procedure; AWS IAM Identity Center for centralized, role-based least-privilege access federated to the institution’s identity provider; AWS KMS encryption at rest and TLS in transit, including the VPN paths; CloudTrail consolidated from three trails to one centralized trail delivered to a dedicated, access-restricted, encrypted log-archive bucket with log-file validation; AWS Config across the estate for continuous drift detection and board-reportable control evidence; and end-to-end validation of the Jack Henry site-to-site VPN. Because the no-console operating model is a genuine workflow change for a two-person cloud team, VeUP stands up a Terraform mock dev/staging environment where the customer’s team practices provisioning and operations before touching production, backed by a multi-AZ disaster-recovery drill against the tiered recovery objectives.

Cutover is explicitly gated and reversible, across six gates: security baseline, connectivity, data integrity, recovery readiness, compliance evidence, and member-impact sign-off. The roughly 800 GB data migration is reconciled source-to-target with integrity checks and tested restores from Amazon S3; no member traffic re-points until every gate passes; workloads cut over one at a time so any failure is contained per workload; and the on-premises estate remains the rehearsed, single-decision rollback target until final sign-off. Modernization is deliberately sequenced after the migration is stable: the SSIS/SSRS pipeline refactors to AWS Glue and AWS Step Functions with Amazon QuickSight on a proven base — the right order of operations for a regulated, always-on banking estate.

The engagement’s posture maps directly onto the six Well-Architected pillars. Operational Excellence: deployment and operational-readiness gaps (OPS06/OPS07) close through the Terraform/pipeline-only production model, mock-environment team enablement, and runbooks drilled before steady state; observability gaps (OPS04/REL06) route to Amazon CloudWatch with Container Insights and AWS X-Ray. Security: the absent incident-response program (SEC10) is remediated with a NIST SP 800-61-aligned IR plan; IAM deficiencies (SEC02/SEC03) with AWS IAM Identity Center, break-glass roles, and CyberArk-brokered privileged access; threat detection with Amazon GuardDuty and AWS Security Hub; and encryption standardized on AWS KMS at rest and TLS in transit. Reliability: the absent DR strategy (REL13) and the single-AZ and single-NAT-Gateway risks the review surfaced convert into multi-AZ deployment for the core tier, tiered RTO/RPO objectives, AWS Backup with cross-region replication, and a rehearsed DR drill. Performance Efficiency: compute is sized against the measured workload rather than carried-forward hardware, validated on the r5 instance family with pipeline-controlled headroom to scale. Cost Optimization: the licensing step-down to license-included SQL Server Standard, the region relocation, transit-gateway collapse, and CloudTrail consolidation all cut standing cost, with AWS Budgets and AWS Cost Anomaly Detection standing guard on the production account. Sustainability: right-sizing an over-provisioned 16-core blade down to 8 cores reduces the estate’s standing footprint alongside its cost.