We moved a global core banking system from an ageing mainframe to AWS one function at a time, keeping both ledgers in sync throughout. Operating costs fell 60%, releases became 5x faster, and there was zero service downtime.
Industry: FinTech & Financial ServicesService: Cloud & DevOpsUpdated
Overview
The challenge
The client's core banking logic ran on a 20-year-old mainframe. New features took 6+ months to release, because any change to the fragile COBOL code could cause failures elsewhere. To cope with holiday peaks, the bank paid for spare hardware in advance, costing millions in idle capacity.
Customer data was reconciled in end-of-day batches, and every mainframe update needed a manual code freeze. The board wanted out, but only if the bank could move without any downtime.
The board judged a single switch-over too risky, because any downtime would bring severe regulatory fines.
MIPS-based pricing from older vendors was eating into the budget the bank wanted for future AI projects.
Experienced COBOL engineers were retiring, leaving fewer people able to maintain the core system.
The solution
We added change data capture (CDC) to stream every mainframe database update into AWS, keeping a live copy of the data in the cloud.
Using the strangler fig pattern, a proxy layer sent specific functions, such as balance enquiries, to new AWS microservices while everything else stayed on the mainframe.
A transactional outbox kept the new AWS ledger and the mainframe 100% consistent throughout the phased rollout.
Before and after
Before
Fixed hardware
Expensive, long-term hardware contracts with little room for sudden peaks.
After
Scales on demand
Cloud services scale automatically and respond to traffic spikes in moments.
Before
Quarterly releases
Mainframe updates needed manual, high-risk code freezes.
After
Daily CI/CD
Automated pipelines let teams update each microservice safely, several times a day.
Before
End-of-day batches
Overnight reconciliation delayed updates to customer data.
After
Live events
Each transaction sends a live event to every system across the bank.
Key features
Updates to the new microservices and the older databases succeed or fail together, which prevented mismatched data during the multi-year migration.
Business impact
100% data consistency
Automated IAM policies and encryption at rest and in transit meet established banking security requirements on AWS.
Business impact
Zero audit findings
Automatic node recycling and circuit breakers stop a single fault from taking down the new core.
Business impact
99.999% uptime
Results
60% Operating Cost Reduction
Switching from mainframe licensing to AWS pay-as-you-go pricing brought immediate savings.
Before: HighAfter: 60% Lower
5x Faster Releases
Product teams now release new savings products without waiting for mainframe code freezes.
Before: 6 MonthsAfter: 4 Weeks
100% Reliable
Because functions moved one at a time, not a single customer session was lost along the way.
Before: < 0.1%After: Zero
Faster delivery
Because these components were already tested, the team skipped repeated setup and moved the first functions off the mainframe sooner.
Gave the team a tested CDC setup for streaming mainframe database changes into AWS.
Tracked AWS spend as each function moved off the mainframe, so savings could be checked against the old MIPS bills.
Monitored the new microservices and the proxy layer, so problems surfaced before customers noticed.
Supplied the outbox logic that kept the AWS ledger and the mainframe in step.
Trust and control
Multi-region AWS residency keeps all banking data within regulatory boundaries.
Compliance readyHardware security modules (HSMs) manage the keys for all personal information and transaction records.
EncryptedAutomated failover and self-healing clusters keep availability at 99.999%.
High availabilityOur project delivery model includes documentation, training and a knowledge handover, so the client's team owns the AWS core on completion.
100% ownedIts core logic ran on a 20-year-old mainframe with fragile COBOL code, so new features took 6+ months. MIPS-based licensing was eating the technology budget, peak capacity meant paying for idle hardware, and the COBOL specialists who maintained the system were retiring. The board would not accept a risky big-bang switch-over.
We used the strangler fig pattern. Change data capture streamed mainframe updates into a live copy on AWS, and a proxy layer sent individual functions, such as balance enquiries, to new AWS microservices one at a time. A transactional outbox kept the new ledger and the mainframe 100% consistent throughout.
Operating costs fell 60% after the move from mainframe licensing to pay-as-you-go AWS pricing. New savings products now launch in 4 weeks instead of 6 months, releases are 5x faster, and the migration ran with zero downtime. Self-healing clusters and automated failover keep the new core at 99.999% availability.
If a big-bang migration is too risky for your board, we can help you plan a phased move to the cloud that keeps every service running.