Legacy Systems in Financial Institutions

Legacy systems remain a defining characteristic of the global financial services landscape. Across banks, insurance companies, and public financial institutions, a significant share of core business operations continues to run on software systems developed decades ago. These systems form the backbone of transaction processing, account management, payments, lending, and settlement workflows, where reliability and correctness are non-negotiable.

The Enduring Role of COBOL

COBOL remains a cornerstone of the legacy landscape in financial services. Despite its age, COBOL-based systems still support a large share of global financial activity, with industry estimates suggesting that over 40% of banking and insurance institutions rely on COBOL for core systems, which process more than 70% of high-volume business transactions and around 90% of ATM operations worldwide. This persistence reflects the nature of financial institutions, where core platforms are designed for long lifecycles, regulatory compliance, and operational stability, meaning they are rarely replaced outright and instead evolve through incremental extensions and integrations, resulting in environments where legacy and modern technologies coexist.

Why Legacy COBOL Systems Are Becoming a Limiting Factor?

Although legacy COBOL systems are widely recognized for their stability and reliability, their role within the financial sector comes with significant constraints. Core banking platforms operate under strict regulatory requirements that demand deterministic behavior, auditability, and absolute functional correctness. While COBOL has historically met these expectations, the long lifespan and rigid architecture of such systems increasingly limit their ability to support evolving business and technological needs.

Over time, COBOL-based and other legacy systems have accumulated substantial technical and structural complexity, resulting in several key challenges:

• Architectural Rigidity: Large, monolithic codebases with tightly coupled components that are difficult to modify or extend.
• Embedded Complexity: Decades of business rules and regulatory logic deeply embedded in legacy code, often with limited or outdated documentation.
• Integration Constraints: Difficulty integrating COBOL systems with modern digital channels, APIs, and data platforms.
• Operational Risk: High sensitivity to change, where even minor modifications can introduce regressions in mission-critical functionality.
• Talent Dependency: A shrinking pool of experienced COBOL engineers, increasing dependency risk, and maintenance cost.

Economic factors further reinforce these challenges. Industry estimates indicate that 70–80% of IT budgets in large financial institutions are still allocated to maintaining existing legacy systems, significantly limiting investment in innovation and future growth. As a result, modernization has become a necessity, driven by the need to preserve stability while enabling long-term sustainability and controlled evolution.

AI-Assisted Modernization of Legacy COBOL Systems

Between 2023 and 2025, LLMs have become a key enabler of COBOL modernization across banking, insurance, and the public sector. Instead of risky full rewrites, organizations use AI to accelerate analysis, automate code transformation, and validate functional equivalence through testing, reducing time, cost, and migration risk.

Key AI Techniques and Outcomes:

• Automated Code Analysis & Documentation: Tools such as IBM watsonx Code Assistant for Z analyze COBOL systems, extract business logic, and generate documentation, cutting analysis time by up to ~94% and improving program understanding by ~79%.
• Automated Code Conversion: Generative AI can translate COBOL into modern languages like Java or C#, but this conversion is not yet reliable enough for institutional use or for highly regulated industries. In such environments, modernization requires a combination of domain expertise and specialized, custom-built AI converters to ensure correctness, auditability, and long-term maintainability of the system.
• Automated Test Generation & Validation: AI pipelines generate unit and integration tests to ensure the modernized system matches legacy behavior, improving regression coverage, and reducing migration risk.

Recent real-world examples confirm these benefits. Major banks such as Citigroup and JPMorgan Chase report strong productivity gains and multi-billion-dollar value from AI-driven modernization, while insurers report up to ~60% effort reduction through automated extraction and conversion of legacy logic. For example, in the public sector, Egypt’s Social Insurance agency achieved a 94% reduction in analysis time and 79% faster understanding of complex COBOL systems using IBM watsonx. At the same time, fintech startups such as Hypercubic and EltegraAI are building AI platforms dedicated to decoding and modernizing COBOL, showing that AI-driven legacy transformation is becoming standard practice rather than an experiment.

Measurable Outcomes of AI-Assisted COBOL Modernization:

• Faster Timelines: Toyota Motor North America used AWS Mainframe Modernization to convert over 40 million lines of COBOL to Java, completing the program about 50% faster and reducing discovery and planning by roughly 75%. These results were achieved with the support of an expert engineering layer that used the tools, steered the transformation, and validated the converted systems.
• Higher Productivity and Lower Costs: Banks such as Goldman Sachs, Citizens Bank, and Citigroup report up to 2× faster task completion and ~20% productivity gains.
• Improved Quality and Lower Risk: IBM and Accenture report ~99%+ functional equivalence in AI-assisted conversions, supported by automated documentation and test generation.
• Mainstream Adoption: Major vendors and consultancies (IBM, AWS, Microsoft, Accenture, Capgemini) now offer dedicated AI modernization platforms, indicating that AI-assisted COBOL transformation has moved into the enterprise mainstream

TIAC’s Initiative

Recognizing these industry trends, TIAC initiated an AI-driven COBOL modernization pilot to evaluate whether such techniques can be applied responsibly in highly regulated financial environments. The initiative was designed as a controlled research effort rather than a production migration, simulating realistic banking constraints while allowing safe experimentation.

Although executed in a simplified test environment, the pilot reflected common financial institution realities, including long-lived COBOL logic and strict correctness requirements. The primary objective was to validate feasibility, correctness, and repeatability, establishing a disciplined foundation for future modernization initiatives that combine AI acceleration with engineering oversight.

Solution

A representative, terminal-based COBOL banking application was developed and used as the legacy baseline for the pilot. This approach was adopted due to limited access to the actual system targeted for modernization, allowing the team to safely evaluate AI-assisted migration techniques in a controlled environment.

The test application was implemented using GNU COBOL with a local relational database, while being guided by assumptions aligned with real-world financial institutions, where enterprise COBOL systems typically run on IBM mainframes, rely on DB2 databases, and operate within z/OS environments. This ensured that the pilot reflected realistic legacy constraints while remaining repeatable and low risk.

The application’s functionality defined the reference behavior against which the modernized system was validated, covering core banking workflows such as

• Account creation and listing
• Deposit and withdrawal transactions
• Balance inquiries
• Loan applications with interest calculations
• Loan status tracking and repayment schedule

Several AI tools were evaluated during the migration process. ChatGPT was primarily used for background research and exploration of COBOL modernization patterns, while Claude Code served as the primary tool for code translation due to its built-in validation capabilities. Google Antigravity was assessed as an IDE-based alternative, with results comparable to direct model usage. All AI-generated outputs were manually reviewed to ensure semantic equivalence.

The migrated application was required to meet a defined set of technical and functional criteria to ensure architectural consistency, testability, and predictable behavior:

Architecture

• Three-tier structure:
  
  • Frontend (React) as the presentation layer
  • Backend (Spring Boot) as the business logic layer
  • Database (PostgreSQL) as the data layer

Core technology requirements

• Java 21 with Maven as the build tool
• Spring Boot 3.3.x for the backend
• React 18.2.x for the frontend
• PostgreSQL as the local database, accessed via JDBC  
• JUnit 5 test suite for backend validation

Outcome

The COBOL modernization pilot successfully demonstrated the feasibility of AI-assisted modernization when applied within a disciplined engineering framework. A terminal-based COBOL banking application was migrated to a modern three-tier architecture while preserving core business behavior and validation logic.

Key outcomes include:

• Successful functional migration of legacy COBOL logic to a modern stack
• Validation that AI tools can significantly accelerate legacy code analysis and translation when combined with structured validation and manual review
• Delivery of a fully working end-to-end application suitable for demonstrations and internal evaluation

This work demonstrates that AI-assisted COBOL modernization is a practical and effective approach when combined with disciplined engineering practices. Industry examples show clear gains in speed, cost, and risk reduction, while the TIAC pilot confirms the feasibility of applying these techniques in a controlled environment while preserving functional correctness. The results indicate that AI can significantly accelerate legacy modernization when paired with rigorous validation and architectural standards, providing a solid foundation for future production-grade initiatives in regulated financial environments.