Changing systems that cannot be switched off

Every organisation of any age has at least one system that everything else depends on, that almost nobody fully understands, and that cannot be taken offline. Replacing it is usually described as a technical problem. It is mostly a sequencing problem.

The failed attempts we see have a pattern. They try to replace the whole thing in one programme, they treat the old system as something to be escaped rather than understood, and they assume the organisation around it will stay still for three years. None of those hold.

What works better is boring. Understand what the system actually does, as opposed to what its documentation claims. Find the parts that can be moved independently. Move the riskiest one first, while you still have the appetite and the budget to recover if it goes badly.

That last point matters more than it sounds. Programmes that do the easy parts first look healthy for eighteen months and then hit the difficult part with no remaining contingency and a board that has stopped paying attention.