We have been brought in to rescue enough ERP implementations to notice that the failure modes repeat. The software is almost never the problem.
Master data is the whole project
Every ERP implementation is a data migration wearing a process costume. If your item master has four spellings of the same supplier, three of them will survive the migration and all three will be used.
Assign an owner per data domain — items, customers, suppliers, chart of accounts — before configuration starts. Not a committee. A person.
"We'll configure it to match our process" is usually wrong
Sometimes your process is a genuine competitive advantage and the system should bend. Far more often it is an accumulated workaround for a limitation of the system you are replacing.
Every customisation is a permanent tax: on upgrades, on support, on the next implementation. Ask of each one whether you would design it this way starting fresh.
The customisations that survive five years are the ones tied to how you actually make money. The rest are fossilised workarounds.
The go-live date is a decision, not a deadline
A date set before the data is profiled is a guess. Treat it as one. The organisations that go live smoothly are the ones that moved the date once, early, on evidence — not the ones that held the date and cut testing.