Your ERP go-live will fail on data, not process

The standard programme plan puts data migration late, after configuration is stable. By then, the discovery that 18% of supplier records have no valid tax identifier, or that three plants have been encoding units of measure inconsistently for a decade, arrives with no schedule left to absorb it.

Profile the source data in the first two weeks instead. Not a sample - full column-level profiling of completeness, uniqueness, referential integrity and format conformance on every object you intend to migrate. It is a fortnight of unglamorous work that reshapes the entire plan, usually by revealing that data remediation is the critical path.

Nobody in a technology team can decide which of two conflicting customer addresses is correct. That knowledge sits with the people who use the record. The failure mode is handing business teams a spreadsheet of 40,000 exceptions and a deadline.

What works is tooling: a queue of exceptions ranked by transactional impact, assigned to the right owner, with progress visible to the programme. Fix the records that appear in live transactions first. A dormant record with bad data can migrate to a quarantine state and be cleaned later - it does not need to hold up a go-live.

Run at least three full mock migrations against production-volume data, each ending in a reconciliation that a finance controller signs. Record counts are necessary but not sufficient: reconcile control totals - trial balance, inventory valuation, open order value - because those are what someone will check on Monday morning.

Reconcile in both directions. Confirming every source record arrived misses records that appeared in the target without a source, which is how duplicate suppliers and phantom stock enter a brand new system on day one.