Which language pair actually pays you best?
Most translators cannot answer that, and it is not a discipline failure: the number is not held in any of the places the work gets recorded.
The question, stated precisely enough to be answerable
Not which pair earned the most last year. That is a volume question and the invoices already answer it. The one worth asking is narrower: per hour actually spent at the desk, after the revision load that pair attracts and the reading that subject demands, which of the pairs you work in pays best? Answer it and the follow-up answers itself, because whatever sits at the bottom of that list is either the pair to raise your rate on or the pair to stop accepting.
Why the invoice cannot tell you
An invoice knows an amount, a date and a client, and honestly that is all an invoice needs to know. It does not record which direction the words went, how long they took, or whether the job ran to a rush deadline. Business tools written for this profession tend to stop at revenue by client, because revenue by client is what an accountant asks for at year end, and revenue by client is precisely the cut that cannot answer this: one agency will send you work in every pair you offer and all of it arrives on the same invoice.
The cut that can
Revenue grouped by source and target language, with the number of projects beside each pair, so the pair that earned most and the pair that needed the most jobs to earn it separate out in one table. Beside that: revenue by specialization, revenue by month, revenue by client together with the share of that client's invoices actually paid, a summary of projects by stage carrying total word counts and average turnaround, and the outstanding balance banded into current, thirty, sixty, ninety and beyond. All of it exports as CSV, which matters because the cut you eventually want is always the one nobody built a screen for.
The hourly figure, and what it is honestly made of
Hours worked are recorded on the project alongside the word counts and the card that priced it, so what a job earned divided by what it took is arithmetic rather than an estimate. That ratio is the one that reorders a client list: a specialization paying a premium per word loses to an ordinary pair you work through at pace, once the reading hours are counted. It has to be said plainly, though. The calculation is only ever as good as the hours entered, and a figure built on hours nobody logged will be confidently wrong. Log them properly on one pair for a quarter before letting the comparison decide anything.
What changes once the number exists
Translators typically discover they have been undercharging by comparing notes with a colleague at a conference, years after the evidence was sitting in their own records. Reading that same evidence quarterly turns a rate conversation from a nerve problem into a preparation problem: you arrive knowing what the work costs you, which pairs subsidize which, and whose volume has quietly stopped being worth its terms. That is the whole return on filling the fields in.