Core Banking Modernization

When a GCC central bank approached Diwansoft, their core banking system was running on 30-year-old COBOL code on IBM z/OS mainframes — the same infrastructure that processed the nation's sovereign treasury operations. The system was stable but strategically paralyzed: 48-hour batch reconciliation cycles, zero real-time reporting capability, and a modernization risk so high that previous attempts had been abandoned after months of failed planning.

The Challenge

The bank's digital transformation mandate was clear: real-time payments, open banking API compliance, AI-powered fraud detection, and mobile-first customer experiences — all while maintaining the 99.999% uptime that sovereign banking demands. The legacy system processed $4B in daily transactions. Any disruption would be measured in economic impact, not IT metrics.

Regulatory requirements from the GCC central banking authority added further constraints: all data must remain within national boundaries, audit trails must be immutable, and any modernization approach must be approved by the IT risk committee.

Our Approach: Augment First, Migrate Selectively

Rather than proposing a rip-and-replace migration that would take 5–7 years and carry enormous risk, Diwansoft designed a three-phase augmentation strategy:

Phase 1 — Intelligent Instrumentation (Months 1–6). We deployed a non-invasive telemetry layer that forwarded transaction events from CICS to Apache Kafka in real time — with zero modifications to production COBOL. This created an event stream that fed real-time dashboards replacing overnight batch reports, ML models trained on 5 years of transaction history, and a feature store enabling sub-40ms fraud scoring on every transaction.

Phase 2 — API Surface Exposure (Months 4–12). We placed an enterprise API gateway in front of CICS transactions, exposing clean REST and ISO 20022-compliant interfaces for mobile banking applications, open banking third-party providers, and internal microservices handling new product features.

Phase 3 — Selective Re-platforming (Months 10–24). Quantitative analysis identified the 22% of COBOL programs accounting for 68% of change requests — these were the migration candidates. Batch reporting, product catalog management, and customer onboarding workflows were re-platformed to cloud-native services, while core transaction processing remained on z/OS where it belongs.

Fraud Detection at Scale

The fraud detection model deployed in Phase 1 became one of the most impactful elements of the engagement. Trained on 5+ years of transaction data across 18M+ accounts, the ensemble model (gradient boosting plus a graph neural network for relationship fraud) scored every transaction in under 40ms.

• False positive rate: 0.3% — versus the industry average of 2–4%
• Detection rate: 94.7% across all fraud typologies
• Zero regulatory findings in two subsequent central bank audits
• $340M in prevented fraud in the first year of operation

Results

The project has since been cited by the regional central banking association as a model for legacy core modernization in sovereign financial institutions.

• 73% reduction in total cost of ownership versus the full replacement alternative
• 400x improvement in reporting latency — from 48-hour batch to sub-second streaming
• 99.999% uptime maintained throughout 24 months of continuous transformation
• Zero production incidents during augmentation, versus the industry average of 12–18% incident rate during legacy migrations
• Time to value: 6 months — real-time dashboards and fraud detection delivered before any migration began