I've found SSIS is great for files with a relative degree of consistency, and has major problems with files that have been mangled by people who don't have the faintest clue what they're doing. What I've done in those cases is pretty much pull the whole thing into a table with 1 column of nvarchar(max) and an ID and nothing else, then process in T-SQL from there.

Didn't have a prior automated process. Before I took it on, it was manual, and took about an hour per file. I took it down, eventually, to about 10 seconds per file (average file size about 5k rows, up to about 20k for the biggest ones). Numbers table and such were key to the process. I could do even better with 2k5 and some of the XML functions, but this was on 2k and didn't have what I would have used.