After a few data conversion projects over the years, I've become convinced that no substantial conversion takes place without rethinking some processes. ("Substantial" generally implying moving from one schema to another which is usually prompted by moving from one system to another.)
Rethinking processes isn't so bad as long as you're expecting it. But for some reason people believe (optimism?) that data conversion is going to be as simple as mapping one field to another, not remembering that the need for the conversion process in the first place was a rethinking of data schema. So nobody allocates sufficient time/money for a solid conversion, the process gets in trouble because everyone is concerned with their little domain and following the holy specifications so the appropriate rear can be covered. Then certain people get called in more than halfway through the process to make things go smoother, but to go smoother would require taking a few steps back and looking at what's really needed but you can't take a few steps back because we're already late and the only direction we can go is forward!
(calm blue ocean, calm blue ocean)
(Originally posted elsewhere)