I would stage the trial in two rounds. First use deliberately awkward sample records to expose mapping and version problems. Then let a small group follow the normal process without reminders. If they create a side spreadsheet or message thread, find out what context the system failed to carry.