How to read a translation quote: rates, minimums, and match discounts

The quote arrives as a spreadsheet: a per-word rate, a few line items you were not expecting, some percentages next to word-count bands, and a total. The rate looks reasonable next to the other quote on your desk, but the totals are far apart, and neither document explains why.

Most of the gap between two translation quotes is not the rate. It is what the rate is applied to, what work is bundled inside it, and what has quietly been left out. Once you can read the structure, comparing two estimates becomes arithmetic rather than intuition.

This article is about reading a quote you have received. Building your own estimate from scratch, before you talk to anyone, is a different exercise with a different starting point.

What is being counted, and in what unit

Per-word pricing sounds like a standard until you ask which words. Source-word pricing counts the text you send. Target-word pricing counts the text that comes back. For a language pair where the target language expands, those are different numbers for the same job, so the same rate is not the same price. Source-based counts also have a practical advantage: you can verify the number before work starts, and the total does not move while the work is in progress.

Japanese and Chinese are usually counted in characters rather than words, because they are not written with spaces between words. A Japanese to English quote might be priced per source character or per resulting English word, and those two bases cannot be compared rate to rate at all. Ask which basis applies, and ask for the actual count the vendor used — not just the price it produced.

Then ask what got counted. Key columns, comment or context columns, do-not-translate entries, markup tags, and duplicated system strings can all be inside or outside the number depending on how the file was analysed. A quote that says one hundred thousand words without saying which hundred thousand is not yet a quote you can check.

The line items that are not translation

Translation is one line. Everything else on the page is process, and process is where two quotes for the same word count diverge. Some vendors bundle these into the rate and some list them separately; neither is wrong, but you cannot compare a bundled rate against an unbundled one without unpacking both.

• Review or revision by a second linguist, sometimes labelled bilingual review — often the single largest optional item
• LQA inside a running build, usually priced per hour or per day rather than per word
• File engineering: parsing your format, protecting tags and placeholders, re-importing, handling anything non-standard
• Graphics or layout work, when text lives inside images or needs typesetting after translation
• Project management, sometimes a percentage of the subtotal and sometimes invisible inside the rate
• Minimum charge, applied per language, per job, or per delivery
• Rush surcharge, triggered below some turnaround threshold
• Voice recording and direction, which is a separate budget rather than a line item

Normalise every quote onto the same sheet before comparing:

Language pair     : JA -> EN
Basis             : source characters / target words  (vendor count: ______)
Translation       : rate x net count after any discounts
Second-linguist   : included / separate line / not offered
LQA in build      : per hour, ___ hours estimated / not offered
Engineering       : fixed fee per delivery / included / hourly
Project management: percentage of subtotal / included
Minimum charge    : per language / per delivery / none
Rush surcharge    : applies under ___ business days
Post-launch batch : rate + minimum per batch, turnaround ___ days

Match discounts and what they assume

If the vendor works with a translation memory, the quote may split your word count into bands: repetitions inside the files themselves, exact matches against an existing memory, fuzzy matches in percentage ranges, and genuinely new words. Each band is charged at a decreasing share of the full rate, and the discount schedule can change the total dramatically for a project with a lot of reused text.

The underlying logic is sound. A segment identical to one that was already translated and approved needs checking rather than translating, and it should not cost the same. But the schedule encodes assumptions that are worth surfacing before you accept it.

The first is ownership. If the memory was built from your previous project, the discount is you being credited for text you already paid for, which is exactly how it should work. If the memory belongs to the vendor and travels with them, the leverage disappears the day you change vendors, and your next quote has no discounted bands at all. That is a procurement question, not a pricing one.

The second is verifiability. Ask for the analysis that produced the bands, not just the totals. It is a standard output of the tools vendors use, and a vendor who cannot or will not produce it is asking you to take the largest variable in the quote on trust.

The third is what actually happens to a matched segment. A one hundred percent match can still be wrong in a new place — the same short English word works as a button label and fails as a stat name, and a line reused in a different speaker's mouth can be the wrong register. If matched segments are billed near zero and never opened, that is precisely where silent errors accumulate. It is reasonable to ask that high-percentage matches still be read in context, and to expect that to cost something.

What the quote does not say

Every quote is silent on a set of questions that move the real total more than the rate you were comparing. All of them can be answered by email in a day.

• Who actually translates: staff, regular contractors, or whoever is available that week — and whether the same people continue on your updates
• Whether any step is machine or AI output with human editing, and whether that step is priced as post-editing or as translation
• How many revision rounds are included before further changes become billable
• What happens when you change source text after work has started, which on a game project is not a hypothetical
• Who owns the translation memory and glossary at the end, and in what format you can receive them
• Whether handling your file format is included, or whether the quote assumes you deliver a clean, flat sheet
• Currency, tax treatment, payment terms, and who absorbs bank transfer fees on cross-border payments

Turning two quotes into one comparison

Start by rebuilding both quotes onto the same scope. Decide the scope you actually want — including or excluding review, LQA, and store text — then ask each vendor to price exactly that, adding back the items they had left out. Vendors expect this and it is a normal part of the process, not a challenge to their pricing.

Compare per-language totals rather than rates. A rate is a component; the total for a language is the thing you will actually spend, and it is the level at which minimums, bands, and bundled process all become visible.

Then extend the comparison from launch to a year. Estimate your monthly volume of new and changed text, apply each vendor's update pricing plus their per-batch minimums, and add twelve months of it. For a live game this frequently reverses the ranking: a vendor with a slightly higher rate but no per-batch minimum and a two-day turnaround can be substantially cheaper across a year than one who wins the launch quote.

Finally, note what you cannot buy back later. Translation memory ownership, glossary handover, and linguist continuity are cheap to secure at contract time and expensive or impossible to obtain afterwards. A quote that is lower because those are not included is not actually lower.

What is fair to ask, and where there is room to move

Asking for a breakdown is normal. A vendor who will not itemise a total is telling you something about how the rest of the relationship will go, and it is better to learn that during procurement.

Flexibility usually exists in structure rather than in the rate. Batching small updates onto a scheduled cadence instead of sending them ad hoc removes repeated minimum charges. Committing to a volume or a term is worth something. Giving a longer lead time removes rush surcharges. Accepting the file format the vendor already handles well removes engineering fees. Each of these lowers your cost by lowering their cost, which is the only kind of discount that survives contact with the actual work.

Pressing hard on the rate itself tends to be counterproductive past a certain point. Below it, the vendor cannot keep the same linguist on your project, and turnover is the most expensive discount you can be given: the replacement does not know your terminology, your characters, or the decisions made in the first delivery, and the inconsistency surfaces in the game rather than in the invoice.

If the total is simply over budget, cut scope explicitly instead of asking for the same scope cheaper. Fewer languages at launch, store page and UI first with narrative in a later wave, subtitles instead of voice — these are decisions you can explain and reverse. A silently thinner process is neither.