Before data moves into a new system, or before a decision rests on it, look at it. Data profiling means checking a real sample against the rules in the data dictionary. The usual findings:
- Duplicates: the same record twice, often from a double submission or a merge.
- Blanks in required fields.
- Impossible values: an accident date in the future, a negative claim amount, an age of 212.
- Values outside the code list: CRASH where only COLLISION is allowed.
- Inconsistent formats: dates written three ways in one column.
For each finding, the business decides a rule: fix at the source, fix during migration, reject, or accept with a note. The BA records those rules as requirements of the migration, with an owner and a count of how many records each affects.