It seems to be much easier to use when dealing with very dynamic documents (variable or partially unknown structure), or when only needing a small subset of a complex document.