11 August 2026

Documentation translation: where the money actually goes

Documentation translation costs more than the per-word rate in the quote. On top of that come the layout rework after delivery, the coordination of reviews in the local subsidiaries and the repetition of all of it at every update. These three items appear on no invoice because they land internally.

1What does the cost actually consist of?

Four items, only one of which appears in the quote. The vendor bills by the word. Layout rework, review coordination and the upkeep of your translation assets land with you.

The fourth item is the uncomfortable one because it never shows up on an invoice. When a reviewer corrects a phrase and that correction only lands in the layout file, the translation memory learns nothing.

In the next document the same passage is suggested wrongly again, corrected again and reviewed again.

A price comparison that looks only at the first item is comparing the smaller half.

2What machine translation removed

Machine translation shortened the translation step from days to minutes. It has not changed the remaining steps.

For technical text, the kind that fills manuals and datasheets, it produces a usable raw draft. What shifted is the starting point of the human work: instead of translating, people now check and correct. That is a different job and a different amount of time, but it is not zero.

Take a workflow with eight steps:

• Prepare the text, have it translated, rebuild it in the layout (these three collapse into one)
• Send the PDF to the local subsidiaries
• Collect the corrections
• Enter the corrections by hand
• Check again
• Sign off

Automation merges the first three. The remaining five stay, and they are what turns days into weeks.

Automation projects speed up the step that was measured and leave untouched the step nobody ever measured.

3What post-editing is, and what it costs

Post-editing is the revision of machine-translated text by professional translators. ISO 18587 describes the process and distinguishes two levels.

• Light post-editing. The goal is comprehensibility. Style and nuance are left alone.
• Full post-editing. The aim is a result comparable to human translation.

For technical documentation, usually only full post-editing is an option. With safety-critical content it is not the model that decides whether a phrase holds up, it is a person with product knowledge.

On price, an estimate with an explicit caveat. In the market, post-editing rates sit noticeably below the rate for human translation, in our observation roughly in the range of half to two thirds. That is an observation from quotes and conversations, not a rate card and not a survey. The actual figure depends on the language pair, the text type, the quality of the raw draft and the required post-editing level.

Either way, a different calculation matters more. A lower per-word rate reduces item one. If items two and three are larger than item one, it barely moves the total.

4Why layout rework never appears in the quote

Because it arises after delivery and is usually not performed by the vendor. Translated text has a different length, and the layout was built around the source length.

The W3C publishes a rule of thumb for how much text grows in translation:

Out of German it works in both directions: English usually gets shorter, Finnish and Polish get longer. For a fixed layout both are a problem, because a caption that suddenly runs 30% short looks just as wrong as one that overflows.

What follows is a series of small jobs that add up:

• Overset text. The frame shows the red plus sign and the rest of the sentence is missing from the PDF.
• Lost formatting. After reimport, text that used to be emphasised is now body text.
• Wrong language attribute. The text is French, hyphenation follows German rules, and the spell check reports nothing.
• Text inside linked images. Labels in graphics are not document text and never appear in the export.
• Cross-references and numbering. Once the breaks move, page references no longer match.

Each one is fixed in minutes. Multiply that by the number of languages and the number of correction rounds, and you see why a manual in fourteen languages takes weeks instead of days.

5What review loops cost

Reviews cost most where the feedback changes file format. The effort is not in the checking, it is in the collecting and the retyping.

The reviewer in the local subsidiary receives a PDF and writes comments into it, or sends an email, or both. Someone then transfers those comments by hand. With fourteen languages that is fourteen sets of feedback in up to fourteen different formats, and every transfer is a place where something can be missed.

A reviewer working in the rendered layout fixes the problem on the spot. A reviewer with a PDF creates a second task for a second person.

The practical difference between a good process and a bad one is therefore not the degree of automation. It is whether the human steps happen inside the layout or outside it.

6Which tool reduces which item

No tool reduces all four items. The useful question is not which one is best, but which one hits the largest item in your case.

The last row is the approach we take, and it has a clear limit: whether a safety notice is technically correct is still decided by a person with product knowledge. That cannot be automated away and should not be.

7Where InTO comes in

InTO is our platform for translating documents inside the layout. InDesign files, including packages with linked media, are uploaded and stay there.

Translation, editing and layout review all happen on the rendered page. Word, PowerPoint and Excel are supported alongside InDesign. Three things change in the workflow:

• The text never leaves the file. That removes the rebuild.
• Approved text goes straight into the translation memory. A correction is already there in the next document.
• Updates only touch what changed. Approved passages are carried forward instead of being rebuilt.

At Hörmann KG, our reference customer, which maintains documentation in more than 30 languages, the result looked like this:

The first two figures come from a layout-heavy InDesign portfolio with DTP previously outsourced in full. A different starting point produces different numbers, hence the conservative range on the right. The detail is in the Hörmann case study.

The split from a project where machine pre-translation was already in use.