Data reconciliation checks that data moved, copied, or transformed between systems still agrees with its source. Teams compare record counts, totals, key values, or individual records between a source and a target, flag the differences, and investigate their cause. Common checks include matching row counts after a migration, comparing daily revenue totals between an ERP system and a data warehouse, and confirming that every order in one system appears in the other.
Reconciliation is a core part of data integrity work in finance, migrations, and data pipelines. For example, if a warehouse shows 10,212 orders for a day and the source shows 10,215, reconciliation finds the three missing records and why they failed to load. The same discipline applies when data is prepared for AI. Comparing the data before and after each preparation step shows what changed and helps explain a change in model results.