Legacy Data Systems Stall Bank AI

Many of the banking industry's core systems run on COBOL, a programming language developed in 1959. Approximately 43% of banking systems are built on COBOL, which still powers 95% of ATM transactions and 80% of in-person transactions. This reliance on decades-old technology creates significant hurdles for integrating modern AI applications. Maintaining these legacy platforms consumes a massive portion of IT budgets, with some estimates suggesting up to 75% of a bank's IT spending goes toward maintenance rather than innovation. The total cost of ownership is often underestimated; one analysis found that the true cost for legacy systems can be 3.4 times higher than budgeted when accounting for inefficiencies and compliance overhead. Global financial institutions spent $36.7 billion on maintaining legacy payment systems in 2022 alone. The structural limitations of these systems, which often operate in silos and process data in batches, are a direct barrier to real-time analytics. AI-driven fraud detection, for example, requires processing enormous volumes of data in milliseconds, a task for which batch-oriented mainframe systems were not designed. This latency can delay threat detection, leaving institutions vulnerable. In response, financial services are forecasted to be the biggest spenders on AI solutions through 2028, accounting for over 20% of all AI spending. Despite the challenges, 80% of banks increased their AI spending for 2025. This investment is increasingly seen as non-negotiable, with JPMorgan Chase reclassifying its $2 billion AI allocation from "discretionary innovation" to "core infrastructure." This technological debt creates a significant competitive disadvantage against fintech startups, which are built on modern, agile infrastructures. Fintechs leverage cloud computing, AI, and machine learning to offer more personalized and efficient services, often with lower overhead costs. This forces traditional banks to accelerate modernization to avoid losing market share. A shrinking talent pool of programmers skilled in legacy languages like COBOL presents another major challenge. With the original engineers of these systems largely retired, banks face higher costs and difficulties in finding qualified personnel to maintain and update the code. One study found that over three-quarters of banks reported having only one or two people capable of maintaining their legacy code. To bridge the gap, some institutions are using "wrappers" or middleware that place a service and innovation layer on top of the legacy core, connecting to modern systems through APIs. This allows for a more gradual modernization, enabling real-time data flows for AI applications without the immediate risk and expense of a full core replacement.