A new report from Ensono puts a staggering figure on a long-standing technical debt: UK banks spend £5.7 billion annually just to maintain legacy systems. For data engineers, this isn't a story about old code, but a system design crisis. Over 90% of these institutions still process core transactions on mainframes, often using COBOL, creating a foundational mismatch for modern data workloads.

The engineering trade-off is clear. These monolithic systems, built for batch processing, cannot provide the real-time, unified data streams required for production AI in fraud detection or customer analytics. Banks are attempting to build event-driven architectures and machine learning pipelines atop foundations designed decades ago. The common 'strangler fig' pattern—wrapping legacy cores with microservices—is architecturally sound but demands years of disciplined execution against constant operational pressure.

Migration risk paralyzes progress. As seen in past failures, a botched core system transition can halt millions of transactions. Consequently, while 67% of banking IT leaders name modernization a top priority, only 28% report substantive progress. The talent gap exacerbates the problem; engineers who can navigate legacy systems are retiring, while those skilled in cloud-native and ML engineering are drawn to more agile competitors.

This creates a direct bottleneck for AI investment. Banks like HSBC and Barclays are committing billions to AI, but executives acknowledge legacy infrastructure as the primary impediment to scale. Clean, governed, and accessible data is a prerequisite for effective AI, and siloed mainframe data is notoriously unfit for this purpose. The result is a costly cycle: budgets allocated for innovation are consumed by the maintenance of systems that actively block it. The banks that solve this architectural puzzle will gain a decisive advantage; those that cannot will see the gap between their ambitions and their infrastructure widen into a chasm.

Source: Webpronews