If you have a data input with multiple headers, you mark them as multiple name rows. In the data matrix that results, the header is a concatenation of the headers, separated by a comma. If you then want to use this header as a category in an unpivot transformation, then split the headers to parse out the different aspects of the rows of the original data source, you have to do it with a comma. However, if a comma is part of the text of the header, it doesn't work. The "|" is a character that does not naturally appear in language (at least in English), so it would be a better character of joining the strings of the name rows.