Every mainframe modernization pitch eventually includes a slide with a clean 18-month bar chart ending in 'legacy fully decommissioned.' Treat that slide with suspicion. Real mainframe systems accumulated thirty years of undocumented business logic, and unwinding that safely takes longer than any fixed-bid quote implies — not because the technology is hard, but because discovering what the system actually does, safely, takes time nobody can shortcut.
A more honest phase breakdown
- Months 1–2: archaeology — mapping data flows, undocumented integrations, and batch job dependencies. This phase produces no visible modernization but prevents every disaster in the phases after.
- Months 2–4: facade and first extraction — an API layer goes in front of the mainframe, and the first low-risk, high-traffic capability gets extracted and parallel-run.
- Months 4–12: incremental extraction — capability by capability, in ranked order, each shipped independently with its own rollback rehearsal. This is the long middle where most of the real work happens.
- Ongoing: the mainframe shrinks, it doesn't vanish on schedule — some capabilities (batch settlement jobs, decades-old regulatory reporting) may reasonably stay on the mainframe indefinitely if extracting them isn't worth the risk.
The mainframe staying partially in place isn't a failure of the modernization — it's often the correct final state. Not every function needs to move; some just need a modern facade in front of them.
Why the honest timeline is the one worth paying for
A vendor promising full replacement in 18 months is either scoping a much smaller system than yours or planning to discover the real complexity mid-project, on your budget. A phased plan that ships value every few weeks, keeps the mainframe as a safety net throughout, and is honest that some functions may never leave, is slower on paper and dramatically less likely to become the multi-year death march that gives modernization its bad reputation.