Translation Quality Is Now a Business Survival Issue: Why “Good Enough” Will Bankrupt Your Global Brand

Updated · 4 min read
On this pageWhat the standards actually defineFour fit-for-purpose levelsTurning levels into numbersWhere undefined quality costs moneyWrite it down in one page

A procurement team asks for "good enough" Korean for a new product line. The vendor delivers lightly edited machine output. The help articles are fine. The warning label is not. Both sides did exactly what they agreed, because "good enough" was never defined per document.

The phrase is not the problem. Spending the same effort on an internal FAQ and a safety label would be wasteful in one direction or dangerous in the other. The problem is leaving "enough" undefined. This post sets out four quality levels, the process each one needs, and a simple way to turn them into measurable acceptance criteria.

What the standards actually define

Three references are worth knowing, because vendors will cite them and they mean less, or more, than people think.

ISO 17100:2015A process standard for translation services. Translators and revisers must meet qualification criteria (a translation degree, another degree plus two years of full-time experience, or five years of full-time experience), and every translation must be revised by a second person. A separate monolingual review is optional.
ISO 18587:2017Covers full post-editing of machine translation and post-editor competences. Light post-editing, which aims only at comprehensible text, appears in an informative annex and is outside the standard's requirements.
MQMMultidimensional Quality Metrics, an open error typology: terminology, accuracy, linguistic conventions, style, locale conventions, audience appropriateness, design and markup. Errors get a severity, usually minor, major or critical.

The first two describe how a translation is produced. Only the third helps you judge a finished text. You need both kinds of reference: a process that makes the right outcome likely, and a measurement that tells you whether it happened.

Four fit-for-purpose levels

LevelTypical contentProcessAcceptance
1. GistReading competitor pages, incoming emails, researchRaw MT or light post-editingReader understands the point; never published or quoted
2. InternalInternal knowledge base, meeting notes, low-traffic support macrosFull post-editing (ISO 18587) or translation with self-checkNo critical errors; score threshold set low
3. PublishedProduct UI, help center, website, marketingTranslation plus revision by a second linguist (ISO 17100), in-context checkNo critical errors, few majors, higher threshold, style guide followed
4. CriticalContracts, labels, safety and medical text, patents, financial disclosuresAs level 3, plus subject-matter review and owner sign-offZero critical and zero major errors; every number verified

Two rules keep this honest. First, the level belongs to the content, not the budget. If a warning label is level 4, it stays level 4 when the quarter is tight. Second, content moves up when it is reused: a level 2 support macro copied into a public FAQ is now level 3 text with level 2 quality. For the difference between light and full post-editing in detail, see MT post-editing guidelines.

Turning levels into numbers

MQM-style scoring counts errors in a sample, weights them by severity, and normalises by word count. A widely used set of default severity multipliers is 1 for minor, 5 for major and 25 for critical. The worked example on the MQM site computes a simple score as one minus penalty points divided by word count, and treats any critical error as an automatic fail. Try it on a sample:

9penalty points
98.2quality score
Passresult

Weights 1 / 5 / 25. Score = (1 − penalty ÷ words) × 100. Any critical error fails. Illustrative model; calibrate thresholds on your own samples.

The useful part is not the number but the conversation it forces. Ask your reviewers to score a sample of Korean you already consider acceptable, and set each level's threshold just below that. Then the vendor, the reviewer and the budget owner are arguing about the same thing.

Where undefined quality costs money

The cost of a vague standard rarely shows up as a line item. It shows up as rework when a level 2 text turns out to be public, as support tickets about an unclear setting, or as a regulator's question about a label. The fix costs more than doing it right would have, because it now involves a release, a reprint or a lawyer. We look at the commercial side in silent attrition in SaaS and at reducing QA cost without lowering the bar in sampling and risk-based QA.

Write it down in one page

  1. List your content types and assign each a level.
  2. For each level, state the process: MT, post-editing, translation plus revision, expert review.
  3. State the acceptance rule: allowed errors by severity and a score threshold.
  4. Name who signs off at levels 3 and 4.
  5. Put the level in every translation request, so the vendor prices and staffs it correctly.

With that page in place, "good enough" becomes a perfectly respectable instruction. For choosing who does the work, see choosing an affordable translation service without compromising quality and AI vs human translation.

Need a quality definition your vendors can work to?

We help teams assign levels to their Korean content and set acceptance criteria, then translate to them. Talk to us →

Work with us

Keep the message. Change the words that need changing.

Korean marketing copy adapted by writers who know the market and your brand voice.

Get a free quote