Replacing a legacy system without stopping the business

Every organisation of a certain age has one: the system everyone depends on, that nobody fully understands anymore, and that everyone is afraid to touch. Replacing it feels like changing the engine mid-flight, which is why these projects get postponed for years. The good news is that a legacy replacement can be made boring, and boring is exactly what you want.

No big bang

The riskiest possible plan is the big-bang switchover: build the new system in secret, pick a weekend, and pray. We do the opposite. The new system grows next to the old one, takes over one slice of the work at a time, and proves itself in production at every step. If a slice misbehaves, you switch that slice back, not the whole company.

The old system is the spec, but not the blueprint

Decades of real use have packed the old system with knowledge: edge cases, business rules, the exception someone added for a reason nobody remembers. We mine it for what must survive, and just as deliberately decide what should not. Recreating every twenty-year-old workaround faithfully is how you build a new legacy system with a modern coat of paint.

Data deserves its own project

Migrating years of data is usually harder than building the new system, and it cannot be an afterthought. Rehearse the migration until it is routine, verify the result against the source, and keep the way back open until the new system has earned trust. Nobody should have to be brave on migration day.

People migrate too

The hardest migration is rarely the data; it is the habits. People have years of muscle memory in the old system, and every one of its quirks is somebody's workflow. So the switch is a change project as much as a technical one: involve the heaviest users early, let the old and new run side by side long enough to build confidence, and treat 'the team prefers the new one' as a launch criterion, not a hope.

The end state we aim for is an anticlimax: the day the old system is switched off and nobody notices, because everything that mattered already moved. That is what a well-run legacy replacement looks like.