Pay-per-document translation: when one-off pricing beats a subscription

Someone in your company has one document and one deadline. A twelve-page supplier agreement in German, due Thursday, and nobody on staff reads German. The question that comes next is almost never about quality. It's about billing: is pay per document translation the right way to buy this single file, or do you put a monthly plan on the company card and stop thinking about it? We get asked this often enough that we've stopped giving a general answer. The arithmetic takes about five minutes, and in our experience almost nobody runs it before buying. Here's how we'd run it.

What pay-per-document translation actually means

Three billing shapes exist in the self-serve market, and the names overlap badly enough to cause real confusion.

The first is a one-off per-word charge. You upload a file, the tool counts the translatable words, you pay for that count once. Most tools attach a minimum charge, so a 180-word birth certificate ends up costing the same as a 900-word one.

The second is a flat rate per document or per page. This is the traditional vendor model and it survives in certified-translation services. A "page" in those quotes usually means 250 words, not a physical page, so one dense table-heavy sheet can bill as three.

The third is a subscription with a monthly word allowance, often with no cap on document count inside that allowance. Unused words usually expire when the cycle resets.

People searching for one-time document translation online generally want the first shape and spend twenty minutes reading pages about the third. And note that per document almost never means per document literally. Word count is still doing the work underneath; the pricing label just hides it.

One more thing about counting, because it trips up budgets in both directions. What gets billed is translatable words, not file size or page count. That includes headers, footers, text boxes, speaker notes in a PPTX presentation, and every visible cell in an XLSX workbook. A 40-slide deck can hold fewer words than a four-page DOCX contract. We've watched a buyer budget 900 dollars for a slide deck that came in under 2,000 words, and we've watched another assume a "short" parts list would be cheap when the workbook held 14,000 cells.

The break-even math most buyers skip

The calculation itself is one line. Divide the monthly plan fee by the one-off per-word rate, and you get the monthly word count where the two models cross.

Say a plan costs 20 dollars a month and one-off work is priced at two cents a word. Break-even is 1,000 words a month. Under that, buying per document is cheaper. Over it, the plan is.

Then two corrections, both of which change the answer more often than the headline number does.

The first is frequency rather than volume. Count the months in which you sent anything at all, not your annual total. A company with 30,000 words in March and nothing for the next five months looks like subscription territory on paper and behaves like a one-off buyer in practice. Annual totals lie about monthly rhythm.

The second is the minimum charge. If the one-off floor is five dollars and your typical job runs 300 words, your effective rate isn't two cents, it's closer to 1.7 cents per word at best and considerably worse on small files. Redo the division with the floor, not the advertised rate, and the break-even point moves.

There's a third factor nobody prices in advance: re-runs. You translate the file, then notice the product names were translated when they shouldn't have been, so you fix the glossary and translate again. Under one-off pricing that second pass is often a second purchase. Inside a word allowance it's just more words off the budget. We'd estimate most first-time buyers re-run at least one file, and anyone working with technical terminology re-runs more.

Worked through: a 4,000-word contract once a quarter is 16,000 words a year spread over four active months. One-off, almost certainly. A 6,000-word HR pack every month is 72,000 words over twelve active months. Plan, almost certainly. The hard cases sit between those two, and the frequency column decides them, not the volume column.

This math does break down in one situation worth naming. If your volume arrives as a single annual spike, say a 60,000-word compliance refresh every January, neither model fits neatly: a one-off purchase that size is expensive and a plan sits idle for eleven months. That case usually needs a vendor conversation rather than a pricing page.

Two cases where paying per document wins

A four-partner law firm we spoke with translates Spanish and Portuguese contracts for one recurring client. Roughly one document a quarter, 3,000 to 5,000 words, always DOCX, and the layout has to survive intact because the file goes back to the client as a tracked redline. Their annual total is around 16,000 words. A 20-dollar monthly plan runs 240 dollars a year to cover four files, and somebody has to remember it exists. Paying for those four jobs individually is cheaper and simpler, and the office manager who buys them doesn't have to own a recurring vendor relationship.

The second case is narrower and comes up constantly. A founder needs one pitch deck translated before an investor trip. One PPTX presentation, once, probably never again. The job also has a sharp constraint: the deck has to come back as an editable PPTX, because the slides will get rewritten on the plane. Nothing about that profile justifies a subscription.

There's a third reason that isn't about price at all. In several companies we've dealt with, a one-time card purchase clears in an hour while a recurring subscription triggers a vendor review that takes two weeks. That's a calendar argument rather than a financial one, and it decides more purchases than anyone puts in writing. If your document is due Thursday and your procurement cycle is measured in weeks, the question of which model is cheaper has already been answered for you.

One-off also wins when you genuinely don't know whether a tool will work on your files yet. Buying a plan to find out is backwards. Buy the smallest unit that gets a real document through the pipeline, look at what comes back, then decide.

When a subscription quietly becomes cheaper

An HR team at a mid-sized manufacturer gave us the clearest version of the other side. They translate safety instructions, onboarding packs, and a parts list that lives in an XLSX workbook. Twelve to eighteen files a month, 5,000 to 8,000 words total, four target languages. Two things pushed them onto a plan, and only one was price.

The price part is obvious: at that volume per-word purchasing costs several times the monthly fee. The second reason mattered more to them. Because their terminology is specific and partly wrong in the first pass, they re-run files. On a word allowance, fixing a glossary entry and translating again costs allowance. On per-document pricing it costs another transaction, another approval, and another line on the card statement, which in practice means people stop re-running and ship the version with the bad term in it.

Glossary continuity is the other thing a plan buys you, and it's underrated. When your projects persist in one place, the glossary you built in March is the glossary you use in November. When you buy scattered one-off translations across whichever tool was convenient that week, terminology drifts, and it drifts in the one word that matters. We've seen the same safety warning come back translated two different ways six weeks apart, which is exactly the failure a glossary is supposed to prevent.

Finance tends to prefer plans too, for an unglamorous reason: one predictable line item beats eleven small card charges that someone has to reconcile.

The honest limitation: a plan helps only if your work shows up every month. If your allowance expires monthly and your documents arrive in two bursts a year, you're paying for ten empty cycles to get a discount on two busy ones. Check whether allowances roll over before you assume otherwise. Most don't.

Costs that don't show up on the pricing page

The per-word rate is the easiest number to compare and the least predictive of what you'll actually spend. These are what we'd look at instead:

• The minimum charge, and what it does to your typical job size rather than your largest one.
• Whether a re-run after a terminology fix costs again, and whether you can even re-run without re-uploading.
• What comes back. An editable DOCX you can work in is a different purchase from a PDF somebody has to retype.
• Whether you get anything besides the translated file. A QA report tells you where the output is weakest. A neutral XLSX export with source and target columns lets you move the content into a CAT tool later, which matters if a professional reviewer gets involved.
• How words are counted, specifically whether hidden content, repeated cells, and speaker notes are billed.
• Seat limits, if more than one person will need to send files.
• Data handling. Whether your files are retained, and for how long, is a question your legal team will eventually ask you rather than the vendor.
• Whether the purchase produces an invoice your finance process accepts, not just a card receipt.

The most expensive hidden cost is rework, and it never appears in a price comparison. If a tool returns a file whose tables collapsed or whose numbering restarted, someone rebuilds it by hand. Ninety minutes of an office manager's time outweighs every pricing difference discussed above. We've covered the broader cost picture in how much it costs to translate a document in 2026, including where agency and freelance rates sit relative to self-serve tools.

Questions worth asking before you pay either way

Short list, and the answers are usually findable in ten minutes of reading or one support message:

• What is the minimum charge, in words or in currency?
• Is the word count shown before payment, or estimated after?
• Does the price include a translated file in the original format, or only extracted text?
• Can I review and edit the glossary and the translation instructions before translation starts, or is it one click and done?
• If I re-translate the same file after a correction, what does that cost?
• What formats are accepted? DOCX, XLSX, and PPTX cover most business documents; scanned PDFs are a different problem with a different price.
• Do I get a QA artifact, or just the file?
• If I later need a human reviewer, can I export the content in a form a CAT tool reads?

The answers that matter most are the third, fourth, and fifth. Format fidelity determines whether you get a deliverable or a draft. Pre-translation control over terminology determines whether the first pass is usable. Re-run cost determines whether you'll actually fix what you find. If you're still weighing self-serve against sending the job out, this comparison of translation bureaus and AI tools for routine business documents covers that decision separately.

How to decide in about ten minutes

Open your sent folder and count. Not impressions, actual numbers.

First, list every document you needed translated in the last 90 days and estimate its word count. Second, count how many of those three months had at least one job. Third, divide a plan's monthly fee by the one-off per-word rate you've been quoted to get your break-even word count. Fourth, compare that break-even against your busiest single month, not your average. Fifth, add one re-run per file to whichever model you're leaning toward and check whether the answer holds.

If fewer than two of the last three months had work in them, buy per document. If all three did and your monthly count clears break-even, take the plan. If you land in the middle, the tiebreaker is re-runs and glossary continuity, and those point to the plan.

For context on where we sit: we're on the subscription side of this ourselves. SnapIntel takes DOCX, XLSX, and PPTX files, runs a preparation step where you review the domain analysis, glossary, and translation prompt before translation starts, then returns the translated file along with a QA report and a neutral XLSX export you can move into any CAT tool. Pro is 20 dollars a month for 60,000 words and unlimited documents. There's also a one-time trial, 5,000 words over seven days, no card, which is enough to put a real document through and see what comes back before any of this math matters.

So here's the concrete step: take your single most awkward document, the one with the tables or the embedded chart labels, and run it through on the smallest unit you can buy. Then look at three things in the output: whether the layout held, whether your product names survived, and what the QA report flagged. That one file will tell you more about the right billing model than any price comparison, because it tells you whether you'll be re-running files, and re-runs are what the models actually differ on.