The translation came back on time. The terminology is right, the agency used its approved glossary, and the reviewer signed it off. The deck still went to a designer for three days.
Answer first: that gap is not a failure of the translation. It happens because translated text changes length and fixed layouts do not. Trados, memoQ, Phrase and Smartcat all export the file back to its original format — what none of them decide is whether the translated sentence still fits the box it has to sit in. The real choice is not which tool your agency holds a licence for. It is whether layout gets resolved by an engine at reassembly, or by a person afterwards.
Text expansion is the constraint, and it is measurable
German runs roughly a third longer than English. Finnish and Russian build compounds that break line-wrapping assumptions. Chinese, Japanese and Korean set to entirely different metrics, so a line that fits by character count can still overflow its shape.
In a DOCX this mostly resolves itself, because the page reflows. In a PPTX, an INDD or a fixed-form PDF, it does not — the shape is a fixed size and the text either fits or sits on top of something else.
This is not a flaw in any one product. DeepL's own documentation, for IDML files, tells users to expect overset text in fixed-size frames, and warns that a target language needing characters outside the original font can render as substituted glyphs or boxes. That is an honest description of what fixed layouts do to every tool that meets them.
Where each tool actually decides layout
Trados Studio is desktop software licensed per linguist, with dozens of format filters and its own SDLXLIFF working format. A file is imported, translated segment by segment and exported back — so layout decisions live in the project setup, and in whoever runs the layout pass afterwards.
memoQ describes itself as a translation management system and a computer-assisted environment for translators. It imports Word, XLIFF, XML, SDLXLIFF and INDD into segments and exports them back to the source format. Same shape: inline tags and filters are the translator's responsibility, whereas the visual result is checked downstream.
Phrase is a headless, API-first localisation platform with GitHub, Figma and Contentful integrations, built for software and content teams. Its unit of work is a string in a repository rather than a document, so page layout is not really the question it answers.
Smartcat pairs 80+ file formats with Adobe OCR on PDFs and a marketplace of around 500,000 linguists. Layout is handled inside the engine, while anything past what the engine produces becomes a paid human editing step.
DeepL covers eleven document formats through its API, including DOCX, PPTX, XLSX, PDF, IDML and XLIFF. Its own documentation notes that PDF translation relies on OCR and can carry a higher error rate, and recommends translating the source file wherever one exists. Pro users can choose to take a translated PDF back as a .docx.
Bluente parses the source file, translates inside its structure and reassembles the result against the original layout, across 27 file types with OCR included. Tracked changes stay tracked with their authorship, comments stay anchored, formulas keep their references and slides keep their masters. Fixed-size frames in an INDD should still get a design review before anything goes to print.
None of these is the right answer for everyone. If your team is localising product strings in a repository, Phrase is built for that and a document engine is not. If you run a production line of linguists working segment by segment and reusing approved translations, memoQ and Trados exist for exactly that. If procurement wants the linguists sourced in the same place as the software, that is Smartcat's model.
The hand-off is what costs the three days
The rebuild is rarely one big failure. It is a hand-off.
One customer described the pattern precisely: the agency does the translation, "but then they put it into a formatted template and they have to sort of copy, paste into those formatted templates." Somebody is being paid to rebuild a document that already existed.
Each pass adds something you did not want:
A second copy of a confidential file, living outside whatever retention policy you wrote.
Version drift, because the layout fix happens on the translated file and never travels back to the source.
Terminology drift, because the glossary lives in the vendor's project file rather than in your workspace. On a 90-page agreement, "agreement" and "contract" both appear and nobody notices until the counterparty does.
Five questions worth asking before the next project
Which of our formats go through without a layout pass? Ask format by format — DOCX, XLSX, PPTX, PDF, INDD. A tool that handles four of the five still leaves you a workflow.
Do tracked changes and comments survive, with their authorship? For anything under review, a flattened redline is a lost audit trail.
Does a PDF come back as a PDF? If the answer is a
.docx, someone on your side is re-laying-out the document.Who owns the glossary? If it lives in the vendor's project, it leaves when they do.
Who checks for overset text, and is it billed? This is the line item the three days hide in.
When a designer is still the right answer
Some files should go to a person, and no engine changes that.
Brand-critical print work in InDesign needs a designer's eye on the spreads regardless of how the text arrived. Anything filed, signed or published should keep a bilingual review step in the process — Bluente measures up to 95% accuracy on legal language against a client's own reference set, and a glossary closes much of the rest, but "much of the rest" is not "all of it" on a document that carries consequences.
The goal is not to remove the designer. It is to stop paying one to rebuild a layout that never needed to break.
See the format-by-format breakdown, including how Bluente compares with DeepL, Smartcat, memoQ and Trados on layout, on the document translation page.