I use so-called Computer-Assisted Translation (CAT) tools to translate documents that are already in editable format (i.e. Word, PowerPoint, etc.). The ones I like best are Wordfast and MemoQ, for which I own a personal license.
CAT tools conveniently split a source file into segments within two columns, with the source language on the left and the target column in which to translate on the right. The right-hand column can be blank or pre-populated with machine translation on the client’s request (in which case the text is post-edited at a cheaper rate).
CAT tools also compute stats about the source text, accurately identifying any repetitions that are often present in balance sheets or user manuals and sometimes even in contracts. These so-called “reps” and “fuzzy matches” are charged at a discounted price compared to the translation rate, based on a percentage charging grid (which is always shared with the client upfront). However, they don’t only afford a sizeable saving but also ensure uniformity of language for chunks of repeated text, that would otherwise be lost if handled by a human alone.