In the screenshot below one can see that with different input files - because they are different e.g. in the fields and data types, the separator used or maybe the encoding - the ETL logic is duplicated. Allthough the same basic logic applies, we need to create multiple transformations because of the file differences.
The result is, that if something changes or the transformation is extended, it has to be done in multiple places. Or if you get additional input files with yet another structure, then you have even more duplication. All this has a bad influence on agility and also quality.
Of course, you could make an effort to generalize some of the logic that all have in common. In this case you would not have multiple (or as many) duplicates. But it makes the overall ETL more complicated because you have things that are different and things that are common. With a growing number of input files (differences) this also gets quickly complicated or even unmanagable.
In this case, when you get additional input files yet in different formates and with different data types, you won't have to touch your ETL logic. You simply define the metadata according to the input file(s) and you are done. So the more different files you have the more you will benefit from this solution.