COBOL Modernization: What Banks Need to Know Before Starting

Why COBOL Modernization Projects Stall

Banks and financial institutions have been running on COBOL for decades. The language works. The problem is that the systems have grown in ways nobody fully planned for, with business logic accumulating across hundreds of programs written by developers who retired years ago and left no documentation behind, or left documentation that no longer matches what the code actually does.

The three failure modes that repeat across modernization projects are predictable once you've seen enough of them. First, undocumented business logic: rules buried inside 30-year-old COBOL programs that nobody knows exist until the new system behaves differently after go-live. Second, retiring developer knowledge: the people who understood the quirks of the system are gone, and what they knew went with them. Third, scope surprises: teams start a modernization project thinking the system is X lines of code doing Y things, and discover partway through that it's twice as large and does things that weren't on any requirements document.

These aren't individual failures of planning or execution. They are structural problems that affect almost every large-scale COBOL modernization effort. The founding team at Replai spent careers at companies including HP, Meta, Microsoft, and Sapiens and kept running into the same pattern: modernization projects stalling because teams spent months just trying to understand what the existing system actually did before writing a single line of new code.

What a Realistic Modernization Approach Looks Like

A modernization project that delivers what it promises starts with a discovery phase, not a migration phase. The two are often conflated, which is where things go wrong. Moving code to a new platform before you understand what the code does means you carry hidden business rules into the new system, or worse, you lose them entirely.

A realistic approach separates three stages. Discovery comes first: map what the existing system actually does, including business rules that were never written down. Then validation: confirm that the discovered logic matches how the business operates today and identify what's still active versus what's been dormant for years. Migration comes last, and it's faster and safer when the first two stages are done properly.

For banks specifically, this matters because the stakes of getting it wrong are high. Loan origination systems, mortgage underwriting logic, compliance rules tied to regulatory deadlines: these aren't the kinds of things you can reconstruct from memory if they get lost in translation. The audit trail matters too. Replai's platform includes the ability to reconstruct historical changes to underwriting logic, such as reviewing modifications made to mortgage underwriting between 2015 and 2020 for a regulatory compliance review, which is the kind of capability that changes how risk and compliance teams think about modernization.

Teams also need to be realistic about what stays and what gets retired. Not everything in a legacy codebase is still in use. Dependency graph analysis that maps which components are actually running in production versus which have gone inactive is how you avoid migrating dead weight.

How AI-Assisted Discovery Changes the Risk Profile

The traditional approach to legacy code discovery is manual: developers read through the codebase, interview anyone who worked on it, and try to produce documentation. It's slow, incomplete, and biased toward what people remember rather than what the code actually does. For a 2-million-line COBOL codebase, manual discovery isn't a phase in the project, it's most of the project.

AI-assisted discovery changes the math. Replai's platform has analyzed tens of millions of lines of code across dozens of modernization projects. The system scans legacy codebases and extracts business rules, validation logic, and dependencies automatically, producing documentation that reflects what the code does rather than what someone thought it did when they last looked at it.

For IT leaders evaluating modernization vendors, the practical question is: how much of the discovery work can be automated before the expensive human effort starts? The answer to that question changes the cost estimate, the timeline, and the risk profile of the entire project. A team that enters the migration phase with a complete map of the existing system is in a fundamentally different position than a team that enters with assumptions.

This matters especially for banks because many COBOL systems have accumulated 50 or more years of business logic with no single person who fully understands the whole. The original authors are gone. Zero developer dependency on legacy knowledge is not a nice-to-have for these institutions, it's a requirement.

What to Evaluate Before You Start

Before committing to a modernization program, IT leaders at banks should be able to answer a few specific questions about their existing systems. How many lines of code are in scope, and has that number been independently verified? What percentage of the business logic is documented, and is the documentation current? Which components are still running in production, and which are candidates for retirement rather than migration?

If those questions don't have good answers, the first step isn't modernization, it's discovery. The projects that stall are almost always the ones that skipped that step.

For teams that want to understand what discovery would actually look like for their specific codebase, the fastest way to get a realistic picture is a direct conversation. Replai works with engineering teams dealing with large legacy codebases and can discuss what a modernization assessment would involve for a system of a given size. You can reach out through the contact page or schedule a demo to see the platform in action with your specific questions in mind.