Two similar records may represent a duplicate or different subjects and products. Before joining them, the actual link must be verified. A quick fix can eliminate important distinctions or make it difficult to reconstruct previous work, especially when records are already used in other documents.
I propose to define which elements make it possible to confirm identity and which require in-depth study. The similarity of the name is a clue, not sufficient evidence. The route must comply with procedures appropriate to the nature of the data and keep the changes made recognizable.
Let's imagine two items with almost equal descriptions but different packaging. An automatic union could alter quantities and orders. A comparison of the references and characteristics allows to distinguish the duplicate from the variant before intervening on the data used by the work.
The final verification checks links to existing tasks. It is useful to know what has been maintained and what has been corrected. Data cleanup gains value when it improves search and preserves the necessary history, avoiding deletions that produce a more difficult problem to reconstruct.