On this page
Why ranking beats coverageScore every content itemWhere cheap is fine, and where it is notSet up so the next language costs lessWhen to upgradeA seed-stage startup sees a steady trickle of Korean sign-ups and decides Korea is worth a real try. The budget for localization is whatever is left after a quarter's cloud bill. The founder's first instinct is to run the whole product, website and help centre through a machine translation engine overnight. The second instinct, after a Korean-speaking advisor reads the result, is to translate nothing until there is money to do it properly. Both instincts are wrong.
On a small budget, localization is a ranking problem. Some words earn their translation cost many times over; most do not, yet. This post is about telling them apart and spending accordingly.
Why ranking beats coverage
Users do not read your product evenly. They read the app store listing or landing page, the sign-up flow, the first screens of the core feature, the price, and the checkout. If those are in good Korean, many users will tolerate an English settings page or an English help article. If they are not, users leave before they reach the parts you translated perfectly.
Demand for native-language content is real. CSA Research's 2020 consumer survey found 76% of online shoppers across 29 countries preferred product information in their own language, with Korea among the highest at 92%. A startup cannot meet that preference everywhere at once. It can meet it on the path every user walks.
Score every content item
List your content in a spreadsheet and give each item three scores from 1 to 3. Multiply them. Translate from the top down until the budget runs out.
| Score | Reach | Moment | Risk if wrong |
|---|---|---|---|
| 3 | Every user sees it | Before sign-up or payment | Legal, money or safety consequences |
| 2 | Most active users see it | During first use | Support tickets, confusion |
| 1 | Few users see it | Later, occasionally | Mild annoyance |
A typical result for a B2C app looks like this:
| Item | Reach × Moment × Risk | Decision |
|---|---|---|
| Pricing, payment and refund terms | 3 × 3 × 3 = 27 | Professional translation now |
| App store listing and landing page | 3 × 3 × 2 = 18 | Professional translation or transcreation now |
| Sign-up and onboarding screens | 3 × 3 × 2 = 18 | Professional translation now |
| Privacy notice | 2 × 3 × 3 = 18 | Specialist translation, legal check |
| Core feature UI | 3 × 2 × 2 = 12 | Machine draft with full post-editing |
| Top five help articles | 2 × 2 × 2 = 8 | Machine draft with full post-editing |
| Settings, admin screens | 1 × 1 × 2 = 2 | Leave in English for now, or machine draft with light review |
| Blog archive, old release notes | 1 × 1 × 1 = 1 | Leave in English |
The numbers are a tool for arguing, not a law. A B2B product with enterprise buyers will score the security and data-processing pages much higher. The point is that the team makes the call explicitly, item by item, instead of translating in file order.
Where cheap is fine, and where it is not
Testing a market
Situation: you are not sure Korea has demandApproach: machine-translated store listing and ad copy, measure clicks and installsGoogle Play offers free machine translation of store listings in some languages and states that those translations are not reviewed by humans. Good enough to test; replace it once the market responds.The first paid screen
Situation: users reach your pricing pageApproach: human translation, prices set in local currency, terms checkedThis is the moment users decide whether to trust you with money. A clumsy sentence here costs conversions you have already paid to acquire.Legal and safety text
Situation: privacy notice, terms, health or safety instructionsApproach: specialist translation, checked by someone who knows local lawErrors here are not just embarrassing. They can create obligations you did not intend or fail requirements you did not know about.Set up so the next language costs less
The cheapest localization decision a startup makes is in the code, not in the translation budget. These steps cost almost nothing early and a lot later:
- Externalize every string. No user-facing text hard-coded in components. Translators cannot translate what they cannot find.
- Use placeholders, not concatenation. "Hello, " + name + "!" breaks in languages with different word order. Korean puts things in a different order from English more often than not.
- Use standard file formats. Common resource formats, or XLIFF, the OASIS standard for moving localizable content between tools, mean any translator or tool can work on them.
- Start a glossary on day one. Twenty product terms with their approved Korean translations prevents the same feature being named three ways.
- Leave room in the layout. Text length changes between languages. Test with long strings before the translations arrive.
- Record which content is machine-only. When budget arrives, you know exactly what to upgrade first.
Post-editing has levels. If you buy machine translation with post-editing, say which level. Full post-editing aims for text comparable to human translation; light post-editing only makes it understandable. ISO 18587 sets requirements for full post-editing. The price difference is real, and so is the quality difference.
When to upgrade
Revisit the ranking when a market crosses a threshold you set in advance: a revenue figure, a share of sign-ups, or the first enterprise customer asking for Korean documentation. Then move the next items up a tier. A Korean startup going the other direction faces the same choices; we cover that in how Korean startups localize outward with generative AI. For leaner experiments, see lean localization experiments, and for the features Korean users expect beyond text, our Korea readiness checklist. When the time comes to decide between engines and people for each document, our AI, human or hybrid decision tree helps.
Localizing into Korean on a startup budget?
We translate the screens that matter most, post-edit the rest to an agreed level, and help you set up a glossary that scales. Get in touch →