Financial Translation and the Small Errors That Cost Real Money
A mid-sized manufacturer once sent an annual report to its new Spanish investors with one number reversed. Not a wrong number, a correctly typed one: 1.250.000 rendered as 1,250,000 in a market where the comma and the full stop do the opposite jobs. The figure was right. The reading of it was off by three orders of magnitude, and the follow-up calls took a fortnight to unwind.
Nobody had done anything careless. The translator was competent, the proofreader was thorough, and neither had been told which convention the target market used. That is the shape of most failures in financial translation. They are almost never dramatic mistranslations. They are small, quiet, procedural, and they surface in front of the people you least want to explain them to.
Accounting vocabulary does not travel
The core problem is that financial terms are not descriptions of universal things. They are labels attached to concepts defined by a particular accounting framework, and the frameworks do not line up.
"Provision" means one thing under IFRS and something meaningfully different in several national systems. "Turnover" and "revenue" are treated as synonyms by most people and as distinct items by accountants in certain jurisdictions. German has a set of balance sheet categories with no clean English counterpart, and a translator who reaches for the nearest-looking word instead of the equivalent concept produces a document that reads beautifully and reports something untrue.
A linguist working on financial documents therefore needs two competences that rarely sit in the same person: real fluency and enough accounting literacy to know when a term is a false friend. The IFRS Foundation publishes its standards in multiple official translations precisely because this problem is hard enough to require an institutional answer.
Where things actually go wrong
Number formatting is the most common failure and the easiest to prevent. Decimal separators, thousands separators, date order and currency placement all vary, and none of them are the translator's default concern unless someone makes them so.
Scale words are the second. The English "billion" now means a thousand million almost everywhere, but "milliard" survives in several European languages, and older documents in Spanish and Portuguese use the long scale. Getting this wrong in a prospectus is not a typo. It is a misstatement.
Then there are the things that are correct but useless. A footnote referring to a domestic tax provision that the foreign reader has never heard of, left untranslated and unexplained, is technically accurate and practically opaque. Good financial document translation adds a short bracketed gloss rather than pretending the reader shares your tax code.
Finally, consistency across a document set. An annual report, its notes, the investor presentation and the press release should use the same term for the same line item. When four different people work on four files in a rush, they will not, and an analyst comparing them will notice.
Confidentiality is part of the specification
Financial material is usually price sensitive before it is public. Earnings figures, merger documents, audit findings and board minutes all carry a window during which a leak is a regulatory event rather than an embarrassment.
This is where the current enthusiasm for pasting documents into whatever chatbot is open becomes genuinely risky. The question of whether it is safe to translate confidential documents with AI is not really about output quality. It is about where the text goes, who retains it and for how long, and whether your provider can answer those questions in writing.
Ask for the specifics. Named translators under signed confidentiality agreements. Encrypted transfer rather than email attachments. A stated retention policy. Machine translation engines that are private rather than public, if machine translation is used at all. Any serious provider of financial translation services will have these answers ready, because their regulated clients ask every year.
Build the glossary before the deadline
The single practice that separates smooth projects from painful ones is agreeing terminology in advance. Not during review, when everyone is tired and the filing date is Friday.
A working glossary for a listed company runs to a few hundred entries. It fixes how each line item is rendered, which house terms stay in the source language, how the entity names appear, and which style guide governs numbers. It takes an afternoon to build the first time and saves days on every subsequent cycle. It also removes the most common source of reviewer disagreement, which is two competent people preferring different correct words.
Pair it with a review by someone who reads financial statements for a living in the target market. Not a second linguist. A person who would notice if the equity section looked strange.
Plan for the calendar, not the document
Reporting seasons are brutally compressed. The gap between figures being finalised and the release going out is often measured in days, and translation gets whatever is left.
Teams that handle this well stop treating translation as a step at the end. They send the stable sections early, the narrative parts that barely change year to year, and reserve the last window for the numbers alone. They brief the translators in advance on what is coming. They keep the same people across cycles so the institutional knowledge stays put.
None of this is exotic. It is the same discipline the rest of the reporting process already applies, extended to the part that reaches everyone who does not read your working language. Given how much of the investor base that now describes, it is a strange place to keep economising.