On this page
The five jobs, whoever does themWhere the hand-offs breakA minimal process that survives release weeksWhat engineering should fix onceVendor, freelancer or in-houseTooling: start small, add when it hurtsA twelve-person startup decides to launch in Korea next quarter. Nobody has "localization" in their title. The product manager assumes engineering will handle the files, engineering assumes marketing found a translator, and marketing assumes the Korean-speaking sales hire will check everything. Three weeks before launch, half the app is in Korean, the onboarding emails are still in English, and nobody can say which version of the pricing page was approved.
That is not a translation quality problem. It is an ownership problem. Lean translation ops means deciding, before the first file goes out, who does what at each hand-off, and keeping the process small enough that a team without a localization department will actually follow it.
The five jobs, whoever does them
Even a one-language launch involves five distinct jobs. In a large company each is a role. In a startup, one person often holds two or three of them. The point is not headcount; it is that every job has exactly one name next to it.
| Job | What it covers | Typical owner at 0–1 localization people |
|---|---|---|
| Localization owner | Scope, deadlines, budget, deciding what ships untranslated, tracking open questions | Product manager or the one localization hire |
| Source and string owner | Externalizing strings, placeholders, screenshots or context notes, merging translated files | One engineer per release, rotated if needed |
| Translator | First translation into Korean, following the glossary and style notes | External vendor or freelance translator |
| Reviser | Second-person check of meaning, terminology and tone against the source | Vendor's second linguist, for critical content |
| In-market reviewer | Does it read like a Korean product? Final sign-off on brand voice and key screens | Korean-speaking team member, partner, or a contracted reviewer |
Two of these are often merged in ways that cause trouble. The first is translator and reviser. ISO 17100, the international standard for translation services, treats revision by a person other than the translator as a minimum requirement. You do not need certification to borrow the principle: whoever translated a contract clause or a price should not be the only person who checked it.
The second is in-market reviewer and translator. A Korean-speaking colleague in sales or support is an excellent reviewer of tone and a fast source of answers. Asking them to also translate a few thousand strings on top of their real job usually means the work slips to evenings and then past the release date.
Where the hand-offs break
Most launch-week failures in a small team happen at one of three hand-offs, not inside the translation itself.
Engineering to translator. Strings arrive as a key-value file with no screenshots and no notes. "Charge" could mean billing or a battery. "Open" could be a verb on a button or a status label. The translator guesses, or sends forty questions and waits. A one-line context note per ambiguous key, written by the engineer who added it, costs a few seconds and saves days.
Translator back to engineering. Translated files come back with a broken placeholder, a missing key, or a string that no longer exists because the source changed in the meantime. If nobody owns the merge, these go live.
Reviewer to release. The in-market reviewer flags fifteen issues in a chat thread. Nobody turns them into tickets, the next update overwrites half the fixes, and the same comments come back a month later. Corrections need to land in the translation memory and glossary, not only in the live file.
The Korean-specific trap in the first hand-off. Korean attaches grammatical particles to nouns, and several common particles have two forms depending on whether the preceding word ends in a consonant or a vowel. When a string reads "Welcome, {name}" or "{file} was deleted", the translator cannot know how the inserted word ends. MultiLingual describes the usual workarounds: choosing a particle that works after either ending, adding an honorific suffix after user names so the ending is predictable, or showing both forms. Tell your translator which placeholders hold user input, product names or numbers, so they can pick the right workaround instead of guessing.
A minimal process that survives release weeks
A small team does not need a localization handbook. It needs a short list that the localization owner can run in under an hour per release. The one below is illustrative; adjust the details to your stack.
- Freeze the source for this batch. Agree on a cut-off. Strings added after it go into the next batch, not into a rushed side request.
- Tier the content. Prices, payment screens, terms, privacy notices and safety text get translation plus second-person revision. Help articles and release notes can get lighter review. Our post on risk-based and sampling-based translation QA covers how to set those tiers.
- Send context with the files. Screenshots for new screens, notes on ambiguous keys, a list of placeholders and what they contain, character limits for tight UI.
- Keep one glossary. Product names, feature names, and the twenty or thirty terms customers see everywhere. Decide which stay in English. A short, maintained list beats a long, abandoned one; see how SaaS teams keep terminology consistent across languages.
- Check the build, not just the file. Someone opens the Korean build and walks the main flows: sign-up, payment, the top five screens. Truncation and line breaks only show up on screen.
- Route reviewer feedback back to the source of truth. Every accepted fix updates the translation memory or glossary, so it is not undone by the next batch.
What engineering should fix once
A lean team cannot afford to fix the same internationalization bug in every release. Four changes, made once, remove most of the friction.
Externalize every user-facing string. Hard-coded text never reaches the translator and ships in English.
Stop concatenating sentences. Microsoft's globalization guidance explains why: fragments arrive out of order, word order differs between languages, and translators lose control of the sentence. Korean places the verb at the end, so an English sentence split into three keys usually cannot be reassembled into natural Korean. Keep each sentence in one string with named placeholders.
Use real plural handling. In the Unicode CLDR plural rules, Korean has a single cardinal category, "other". Code that appends an English "s" based on a count is harmless for Korean but wrong elsewhere; using your framework's plural support from the start keeps the next language cheap.
Run a pseudolocalized build. Pseudolocalization replaces the English text with accented, expanded and delimited versions, so truncation, concatenated fragments and hard-coded strings become visible before any translator is paid. Microsoft recommends using it early and throughout development, while noting that it does not replace testing the real translated product. For app-specific checks, see our guide to Korean translation for software and apps.
Vendor, freelancer or in-house
At zero or one localization person, the decision is less about cost per word than about which jobs you can staff reliably.
Many small teams end up with a mix: an external translator or vendor for the first draft and revision, an internal reviewer for voice and key screens, and machine translation with human post-editing for low-risk, high-volume text. Our post on hybrid AI translation pipelines explains where that split works. Whoever you choose, assess them the way you would a key hire; choosing a Korean translation partner lists what to ask.
Tooling: start small, add when it hurts
For a first market, a shared glossary, a repository-connected translation platform or a disciplined file exchange, and a ticket label for localization issues are enough. Add a translation management system when manual file copying starts causing missing keys or overwritten fixes, and add automation when releases become frequent enough that the localization owner spends more time moving files than answering questions. Tools should follow pain you have already felt, not pain you expect.
If the Korean launch is still a test, keep the scope narrow: translate the screens and pages that decide whether a Korean customer signs up and pays, measure, then expand. Our post on lean localization experiments shows how to size that first step.
Need a translator and reviser who work from your repository and answer to one owner on your side? See our translation services or contact us with your release schedule.