Answer
First, sources and overlaps are inventoried: ERP, shop, spreadsheets, file shares, supplier data and manual texts. This is followed by a target model with clear field meanings. Duplicates, units, categories, attribute names and variant relationships are cleaned up on the basis of rules and samples.
Full automation rarely makes sense. Rules can handle many cases, but products that are unclear in business terms need a clarification list with a responsible person. Test imports show which information was lost or wrongly assigned. Before the final switch, quantities and critical examples are reconciled. Historical files can stay in an archive if structuring them completely brings no economic benefit. The effort depends heavily on the quality of the source data and should not be estimated from the number of records alone.
As of 3 September 2026
Further reading
Related topic: Product data