Answer
A rollback path only works if it is planned before the change, not improvised after something breaks. On IBM i that often means keeping the existing 5250 screens available in parallel with a new interface for a defined cutover window, relying on journaling so data changes can be traced and reversed, and taking an LPAR-level backup or storage snapshot immediately before a cutover so the team can restore a known-good state within a defined recovery window rather than debugging live.
Piloting with a real workflow, not a demo dataset, is what actually proves the plan. Pick one warehouse, one product line, or one accounts payable process, run it end to end with real transactions and real users, and log every assumption that turns out to be wrong. Those findings are cheap to fix in a pilot and expensive to fix after a full rollout. Buyers should also ask prospective partners how they handle a failed phase in a live client engagement, since the answer says more about their real discipline than any reference story about a successful one.