Migrating off legacy systems
You move off a legacy system without a big-bang cutover, by running the new path beside the old one.
A migration should change the platform and nothing else. Same numbers, same reports, cheaper and faster underneath. The way to guarantee that is to never be in a position where you have to believe it: the new pipeline runs beside the old one, produces the same outputs, and gets compared until the diffs go quiet.
Traffic moves piece by piece. The tangled parts get re-modeled on the way, storage and compute get tuned for what queries cost, and the undocumented parts finally get written down. The old system is retired only when nothing reads from it.
Two warnings from experience. The old system runs longer than anyone plans, so budget for the overlap. And somewhere in the legacy code there is one report whose numbers have been wrong for years. The new platform will expose it, and that conversation is part of the project too.