A Taichung industrial parts manufacturer had 8,000 monthly sessions on its website. About 1,100 came from Vietnam, Indonesia, and Germany — but those visitors averaged just 21 seconds on site. The reason was simple: the entire site was in Traditional Chinese. Overseas buyers skimmed the homepage through browser auto-translation, never found the spec sheets, and never understood the quote button. Three months after adding English and Vietnamese versions, the same traffic produced 37 qualified inquiries. A multilingual site is never just "translating the text" — it reshapes your URL structure, database fields, search indexing, and payment setup all at once.
When It Fits — and When It Does Not
Good time to start:
- Non-Chinese sessions already exceed 8% in GA4, with real inbound emails to match
- B2B manufacturing or trading, where overseas buyers search part numbers in English
- Cross-border commerce formally entering Japan, Korea, or Southeast Asia
- Overseas trade shows that require localized landing pages and catalog downloads
- A SaaS or tool product opening up international signup and subscriptions
Do not do it yet if:
- Your customers are 100% local services (neighborhood restaurants, local tutoring, on-site repair)
- Content changes weekly and nobody owns maintaining N translated versions
- You want an English page "to test the water" but have no one who can reply in English
- Budget is under NT$60,000 but the scope demands four languages
- The product lacks regulatory certification in the target market, so you cannot ship anyway
Alternatives Matrix: Four Ways to Get There
| Approach | Cost range | Strengths | Deal-breaker |
|---|---|---|---|
| Browser / Google Translate widget | NT$0 | Free, live the same day | Translated content is never indexed — zero international SEO value |
| Proxy SaaS (Weglot, Localize) | ~NT$900–9,000/month | Live in 1–2 weeks, auto-detects new strings | Word caps, URLs tied to the vendor, painful to migrate away |
| WordPress WPML / Polylang | NT$3,000–15,000/year + setup hours | Fastest route for content sites, familiar admin | Plugin conflict risk; struggles with deep data structures like spec tables |
| Native i18n build (Laravel / Next.js + next-intl) | From NT$180,000 | Full control over URLs, caching, SEO, and payments | Higher upfront cost; needs a clear language-priority plan |
A simple rule: if the overseas revenue target exceeds NT$3M/year, or your product data is more than three levels deep, go native i18n. Otherwise validate demand with SaaS first.
The Full Process: Phases, Timing, Deliverables, Tools
- Phase 0 | Market and language audit (3–5 days): pull country traffic and query terms from GA4 and Search Console. Deliverables: language priority matrix and a URL strategy decision memo (subdirectory /en/ vs subdomain en.). Tools: Google Analytics 4, Search Console, Notion.
- Phase 1 | Content inventory and string extraction (5–7 days): crawl the site for pages to translate and hardcoded copy. Deliverables: content inventory and i18n key map. Tools: Screaming Frog, Google Sheets.
- Phase 2 | Architecture and design (7–10 days): deliver the hreflang spec, language switcher design, and layout rules for long strings and RTL (Arabic, Hebrew). Tool: Figma.
- Phase 3 | Development (3–5 weeks): locale routing, multilingual CMS fields, and currency/number formatting via Intl.NumberFormat. Tools: Laravel / Next.js, next-intl.
- Phase 4 | Translation and review (2–4 weeks, parallel with dev): machine translation as a base plus native-speaker review; build a translation memory and glossary. Tools: Crowdin, Google Cloud Translation.
- Phase 5 | Testing and launch (5–7 days): bidirectional hreflang validation, Search Console international targeting, locale-aware cache partitioning. Tools: Screaming Frog, Cloudflare.
Real Cost Breakdown (TWD)
- Bilingual retrofit of an existing site: NT$120,000–200,000
- Full internationalization (4–6 locales + translation workflow): NT$250,000–450,000
- Each additional locale: NT$25,000–45,000 (strings, content fields, hreflang, QA)
- Human translation: roughly NT$1.5–4.5 per character (ZH→EN); native review NT$0.8–1.5 per character
- Machine translation API: Google Cloud Translation is about US$20 per million characters with the first 500,000 characters each month free — a first full pass on a 100k-character site often costs NT$0–200
- Localization platform: Crowdin Pro around US$59/month (~NT$1,900)
- Cross-border payments: Stripe adds 1.5% for international cards and another 1% for currency conversion, pushing the effective rate close to 5.4%
- Font licensing: Japanese, Korean, and Arabic web fonts NT$0–30,000/year
- Annual maintenance: 15–20% of build cost (new content translation, locale sync, hreflang monitoring)
What Clients Expect vs. What Actually Happens
| The expectation | The reality |
|---|---|
| Translation just means handing files to an agency | About 40% of the hours go into extracting hardcoded strings and fixing layouts |
| Adding a locale means writing one more copy of the content | Every future release must sync across N locales — maintenance scales linearly |
| Overseas orders arrive the day after launch | Full hreflang interpretation and indexing typically takes 4–12 weeks |
| The English version can reuse the Chinese layout | German strings run 30%+ longer, breaking buttons and navigation everywhere |
| One payment provider covers every market | Japan favors convenience-store payment, Southeast Asia favors e-wallets — separate integrations |
Seven Fatal Traps and How to Avoid Them
- Hardcoded strings scattered everywhere: run a string scan before development, force all UI copy into language files, and fail CI on untranslated keys.
- Asymmetric hreflang: if A points to B but B does not point back, Google ignores the whole cluster. Validate bidirectionally with Screaming Frog and add x-default per Google Search Central's localized-versions guidance.
- Automatic IP-based redirects: they block crawlers and annoy users. Show a switch banner instead and remember the user's choice.
- Switching language by cookie on one URL: search engines only ever see one language. Every locale needs its own indexable URL.
- Translating body copy only: meta titles, descriptions, OG captions, PDF catalogs, and text baked into images get missed. Put SEO fields into the translation key list from day one.
- Hardcoded dates and currency: stop baking 2026/08/06 and NT$ into templates — output them per locale via the Intl API.
- Launching six languages at once: quality and process collapse together. Run one locale end to end first, then scale in batches.
Success Metrics and the First 90 Days
- Day 30: Search Console index coverage ≥ 80% per locale, zero hreflang errors, and LCP under 2.5 seconds on every locale; check language switcher click rate.
- Day 60: organic traffic share and bounce rate per locale, missing-translation rate below 2%, and a first comparison of inquiry/add-to-cart behavior by country to retune landing copy.
- Day 90: calculate conversion rate and cost per lead by country, decide the second wave of locales; translation memory typically saves 20–35% on repeated word volume by this point.
Decision Checklist: 12 Questions First
- ☐ Has overseas organic traffic exceeded 8% over the past 90 days?
- ☐ Have you received at least three foreign-language inquiries?
- ☐ Is there someone who can reply to customers in that language?
- ☐ Are the target market's certifications already in place?
- ☐ Have you settled the URL strategy (subdirectory vs subdomain vs ccTLD)?
- ☐ Is your product data more than three levels deep (category / model / spec)?
- ☐ Does each locale have a named content owner?
- ☐ Have you listed the SEO fields that must be translated?
- ☐ Do you need RTL (Arabic / Hebrew) support?
- ☐ Are cross-border payment and logistics options confirmed?
- ☐ Does the annual budget include translation and maintenance (15–20% of build cost)?
- ☐ Are you willing to launch only 1–2 locales in the first wave?
Eight or more "yes" answers usually mean the ROI holds up. Fewer than five, and you should validate demand with a SaaS approach first.
FAQ
Subdirectory or subdomain for a multilingual site?
For most SMEs, subdirectories (example.com/en/) win. They inherit the main domain's authority, certificate, and CDN configuration, and cost the least to maintain. Only consider subdomains or country TLDs when each market has its own team, product line, and hosting requirements.
Can we use machine translation only?
Use it as a base, never as the final output. The practical pattern is machine-translating the whole site, then having native reviewers correct product names, regulatory wording, and marketing copy. The homepage, product pages, and quote flow must be human-reviewed; the rest can be handled in batches.
Will a multilingual site trigger duplicate content penalties?
No. Different-language versions count as localized variants, not duplicates, as long as hreflang annotations are correct and reciprocal. The real risk is same-language, different-region versions (en-US vs en-GB) with near-identical content — those need genuine localization differences.
How long does the whole project take?
A bilingual retrofit runs 6–8 weeks; a full 4–6 locale internationalization runs 10–14 weeks. Translation and development can run in parallel, but string extraction must finish before translation starts or you will pay for heavy rework.
What does monthly maintenance cost after launch?
For a mid-sized four-locale site: about NT$1,900/month for the localization platform, NT$3,000–12,000/month for new content translation, plus technical upkeep — roughly NT$8,000–20,000 per month in total.
Next Step
ScriptWalker's Multilingual Website Internationalization package (from NT$180,000) covers language-priority auditing, URL and hreflang architecture specs, translation workflow setup, and 90-day post-launch performance tracking. We offer a free 30-minute locale feasibility review using your own GA4 and Search Console data: which locale to launch first, an estimated first-year total cost, and a checklist you can execute yourself.
- Email: [email protected]
- Phone: 0916-224-047
- LINE: @ufv9089p