How to price a translation from a Trados, memoQ or Wordfast analysis

When a job comes with a CAT tool analysis, the price no longer comes from multiplying every word by your rate. It comes from the weighted word count: each band of translation memory matches is paid at a different percentage. This guide explains what each band means, what each program calls it, how the sum works and what to check before accepting an agency's discount grid.

1. What an analysis is

The CAT tool splits the source text into segments and compares each one with the project's translation memory. Depending on what it finds, the segment falls into a band: no match, a fuzzy match in one of several percentage ranges, an exact match, a context match, or a repetition within the job itself. The analysis report adds up the source words in each band.

Two practical consequences:

• The analysis depends on the memory it is run against. The same text gives a different breakdown with a different memory, and the memory is usually the agency's or the client's, not yours.
• A fuzzy match percentage measures how similar the segment's text is to the one in memory, not how much work it takes. An 85% match may need one figure changed or half a sentence rewritten.

2. The bands and what each program calls them

All three programs split the words into the same broad bands, under different names. The fuzzy match ranges they ship with are 95–99%, 85–94%, 75–84% and 50–74%; they can be configured, so a report may come with others. Below the lowest range, the segment counts as no match.

The names are the ones in each program's English interface. Each program also has rows of its own that are worth knowing how to read:

• Trados — PerfectMatch: segments that match an earlier bilingual version of the same document. The discount grid should say at what percentage they are paid.
• Trados — Locked: locked segments, which only appear if the analysis was run asking for them separately. Whether they are paid is the client's call.
• Trados — internal fuzzy matches: segments similar to others in the same job that are not in the memory yet. They only appear if enabled in the analysis; check in the report whether they come separately or inside the new words.
• memoQ — X-translated: segments pre-translated with X-translate from an earlier bilingual version of the same document. The discount grid should say at what percentage they are paid.
• memoQ — Fragments: words memoQ covers by assembling pieces from the memory or the termbase. They are not whole-segment matches, so there is no obvious way to pay for them: agree it with the client.
• Wordfast — Ignored: segments that were excluded from the analysis.
• Wordfast — Leveraged, Internal Fuzzies and Fuzzy Repetitions: these are not bands but subtotals of the others. Leveraged adds up the context, 100%, fuzzy matches and fuzzy repetitions, and the fuzzy repetitions are already inside Repetitions. Adding those rows to the rest counts the same words two or three times.

3. The weighted word count

Each band is paid at a percentage of the full rate. The words in each band, multiplied by its percentage, give the weighted words; their sum, times the per-word rate, gives the price:

price = Σ (words in band × band percentage) × per-word rate

An example with a 5,000-word job, at €0.09 per source word and with the discount grid from the next section:

3,405 weighted words × €0.09 = €306.45. Without discounts, the 5,000 words would have come to €450.00: the grid takes €143.55 off the job, almost 32%. That is why the discount grid matters as much as the rate.

4. The discount grid

There is no official or mandatory reference grid: each agency applies its own, and a translator working with direct clients can propose theirs. The one in the example above is the one Gestrics ships with, and it is only a starting point:

What matters when comparing grids is not each percentage on its own but the result on the real breakdown of your jobs: the same grid is generous with an almost entirely new text and very harsh with an update full of repetitions.

5. What to check before accepting an agency's grid

• Whether 100% matches and repetitions are paid. Even when the memory proposes the translation, it has to be read in place and checked to fit. 0% on those bands is unpaid work.
• What is expected of context matches. If they are paid at 0%, ask whether they have to be reviewed. If they do, that is working time and it needs a price.
• The report, not just the figure. Ask for the analysis file: with it you can check the breakdown band by band and that the bands add up to the total. With Wordfast, check that nobody has added up the subtotals.
• The memory the analysis was run against. If it is old or of uneven quality, a 100% match may need more correction than a fuzzy one. Worth knowing before you accept.
• The rows that are not bands. Locked (Trados), Fragments (memoQ) or Ignored (Wordfast): agree whether they are paid and at what percentage.
• A minimum fee per job. On a job of a few words, the weighted count can fall below what it costs to handle it. A minimum fee per job prevents that.