Almost every business we have worked with keeps one physical register that nobody wants to give up, even after a new system arrives. It is usually a notebook or an Excel file kept by one specific person, and it survives because it does something the new system does not: it is fast to write in, it never goes down, and it matches exactly how that person thinks about the work.
The mistake is treating this as a training problem. It is a design problem. If entering something into the new system takes longer, or requires information the person does not have at that moment, the register wins every time, no matter how good the reports on the other side look.
What actually gets a register retired
We have only ever seen a register genuinely disappear when the replacement was faster to use for the single most common action, not for the tenth most common one. That usually means fewer fields, sensible defaults, and letting the exception be entered messily rather than blocking the whole entry until every field is perfect.
Run the new system beside the register for a real stretch of time, not a demo. When the two start agreeing on their own, the register goes. Not before.