Japan’s core banking and insurance systems still run on COBOL and mainframes, and the engineers who understand them are retiring faster than replacements arrive. Japan’s Ministry of Economy, Trade and Industry (METI) named this problem years ago, the “2025 Digital Cliff,” warning that continued legacy dependency could cost the economy as much as 12 trillion yen a year. For the finance and insurance firms Ahead works with, the sharpest edge of that cliff is the workforce rather than the code: the question is who is left to run these systems.
1. The numbers describe a workforce shortage
While the technology itself remains stable, the workforce that keeps it running is thinning. Japanese industry data indicates that the majority of COBOL engineers in Japan are over the age of 50, while only a small fraction of new engineers enter the field with any COBOL experience at all. The broader shortage compounds it: METI’s IT human-resources survey projects a gap of up to roughly 790,000 IT professionals by 2030 under its higher-demand scenario, with a floor of around 410,000 even under conservative assumptions, and IPA, the agency that maintains this data, places the legacy skill sets among the hardest to fill. For a bank or insurer, that shows up as longer turnaround on changes, rising labor costs, and deepening dependence on a small group of specialists.
2. The knowledge gap runs deeper than the language gap
Learning COBOL syntax is the manageable part of the problem, and the training programs that exist can address it. The greater difficulty is that these systems were built and modified over decades by engineers who left little documentation behind, leaving the business logic distributed between the code itself and the institutional memory of the people who maintained it. Japanese industry surveys reflect this directly, with a substantial share of organizations citing loss of system knowledge as a leading barrier to modernization. The result is a black-box effect: as the people who understood a system depart, that understanding departs with them. The scarce resource is therefore not a COBOL coder, but an engineer who understands what a specific institution’s system does and can modify it safely.
3. AI handles the code; people handle the judgment
Modernization vendors increasingly use large language models to document undocumented code, map dependencies, and assist COBOL-to-Java conversion, and those tools genuinely compress the slowest phases of a project. What they leave untouched is the judgment that matters most in a regulated core system: deciding what is safe to change, in what order, and how to confirm a converted component behaves identically to the original under real transaction load. In banking and insurance that decision carries regulatory and customer-facing weight, and it stays with experienced engineers however capable the tooling becomes. AI narrows the coding gap, but the knowledge-and-judgment gap it leaves open is the one that determines whether a modernization succeeds.
4. How firms should staff the problem
A pure hiring push runs straight into a shrinking, aging talent pool at rising rates. The more durable approach pairs a small number of engineers who understand the legacy system’s business logic with modern engineering judgment and AI-assisted tooling for the work that does not require that understanding. That reframes the staffing question. The scarce skill is the combination of legacy-system domain knowledge and the judgment to sequence a modernization safely, validate AI-assisted output, and own the go/no-go decision at each cutover, rather than raw COBOL maintenance capacity on its own.
Where Ahead fits
Solving this comes down to the team a firm can put on the project, which is where Ahead Group works with organizations across Japan’s finance and insurance sector. For a modernization program, the answer is rarely a single hire. It is assembling the team the project needs: the few engineers who genuinely understand legacy cores, strong general engineers to absorb the domain knowledge and carry the delivery, and AI-assisted tooling for the mechanical work. Because Ahead’s own team comes from technical backgrounds, it can evaluate candidates for that mix rather than screening on a keyword. If your organization maintains a mainframe or COBOL core and is planning the project that comes next, learn more about our Staffing Services.