Legacy Data Systems Stall AI in Banking
Banks are struggling to scale AI initiatives because of real-time data gaps in their legacy IT environments. A new report from Info-Tech Research Group finds that while AI is being adopted for fraud detection and risk analytics, structural data limitations are a major roadblock.
Maintaining legacy systems can consume 60-80% of a bank's IT budget, with some analyses suggesting the true total cost is 3.4 times higher than budgeted when factoring in inefficiencies and missed opportunities. These systems, often decades-old mainframes running on languages like COBOL, still process the vast majority of credit card transactions and are used by over 90% of financial services firms. The fundamental roadblock for AI is the architecture of these legacy systems, which were designed for batch processing rather than the real-time data streams modern AI models require. This creates data silos, quality inconsistencies, and an inability to integrate with modern APIs, hindering the development of sophisticated AI for anything beyond static reporting. Two-thirds of banking leaders liken running AI on these systems to "fuelling an EV with petrol." To bridge this gap, standardization is becoming critical. The ISO 20022 messaging standard, with a final adoption deadline of November 2025, is creating a richer, more structured data format for cross-border payments, which is a natural fit for AI analytics. Concurrently, ISO 42001 has emerged as the first international management system standard for AI, providing a framework for governance, risk management, and compliance. The regulatory landscape for AI in finance is diverging globally. The EU has implemented its risk-based AI Act, which classifies many financial applications like credit scoring as "high-risk," imposing strict transparency and oversight rules. In contrast, the U.S. follows a sector-specific approach with agencies like the SEC and CFPB applying existing rules, while China employs a centralized, security-driven model with its own Interim Measures for Generative AI. This technological lag occurs as AI-driven fraud becomes a significant threat. In 2024, AI-driven tactics constituted 42.5% of all detected fraud attempts in the financial sector. While over 90% of financial institutions are now using AI to combat fraud, only 22% have implemented dedicated AI defenses, exposing a critical vulnerability.
