Hmm I certainly see what you are getting at. The only problem with this is every columns length is different so It wont be so much a search and replace on the length, but going through and changing each one (back to where we started). Unless I am missunderstanding you (ok, might be able to do some funky stuff in excel or sql to generate the correct xml, as we have the data dictionary from the 3rd party).

I have already tried a conversion step in the hope it might take the target's datatypes somehow. One of the benefits of not using the conversion is that at least all my source and tartget columns map automatically by name!!

Feels like any method is extreemly manual, and seems odd that SSIS doesnt have the functionality to cope with large scale imports.

My last resort is the "suggest data types" option but our sample files currently only contain 10 rows which is not representative of the production files (we wont get those til later). My concern is using a type too short for the source which would then cause the data to actually be truncated rather than just warning of it.