Circulos AI

Support AI Skill

Multi Language Knowledge Base

Build and maintain a knowledge base in multiple languages to serve global customers effectively. Use when translating help content, managing multilingual articles, ensuring translation quality, localizing support content, or measuring multilingual self-service effectiveness. Triggers on phrases like "multi-language KB", "translated knowledge base", "multilingual help center", "content translation", "localization support", "global knowledge base", "translation workflow", "language support", "i18n help content", "multilingual articles".

Multi-Language Knowledge Base Management

Build and maintain support content in multiple languages — ensuring global customers receive accurate, culturally appropriate, and timely help in their preferred language.

Workflow

  1. Identify target languages based on customer distribution and ticket volume.
  2. Choose translation strategy: professional, community, AI-assisted, or hybrid.
  3. Set up language-specific content structure and workflows.
  4. Translate priority articles first (based on search volume and impact).
  5. Implement translation quality assurance (accuracy, cultural appropriateness).
  6. Establish maintenance workflow for ongoing updates and new content.
  7. Measure multilingual KB usage and deflection effectiveness.
  8. Expand language coverage based on demand and ROI.

Language Prioritization

LANGUAGE PRIORITIZATION MATRIX
================================

Step 1 — Customer Distribution Analysis:
  ════════════════════════════════════════════════════════════════════════
  Language  | Customers  | % Total | Tickets/Mo | Priority Score
  ════════════════════════════════════════════════════════════════════════
  English   | 50,000     | 50%     | 2,500      | P0 (baseline)
  Spanish   | 15,000     | 15%     | 1,200      | P1
  French    | 10,000     | 10%     | 800        | P1
  German    | 8,000      | 8%      | 600        | P1
  Japanese  | 5,000      | 5%      | 400        | P2
  Portuguese| 5,000      | 5%      | 400        | P2
  Chinese   | 3,000      | 3%      | 300        | P2
  Other     | 4,000      | 4%      | 300        | P3
  ════════════════════════════════════════════════════════════════════════

Step 2 — Priority Scoring Formula:
  → Priority Score = (Customer %) × 0.4 + (Ticket Volume %) × 0.4 + (Growth Rate) × 0.2
  → P1: Translate top 200 articles within 90 days
  → P2: Translate top 100 articles within 180 days
  → P3: AI-only translation or English-first approach

Step 3 — Translation Phasing:
  Phase 1 (Month 1–3): English → Spanish, French, German (top 200 articles)
  Phase 2 (Month 4–6): Add Japanese, Portuguese (top 100 articles)
  Phase 3 (Month 7–9): Add Chinese, expand existing languages (top 300 articles)
  Phase 4 (Month 10–12): Add remaining P3 languages, optimize all languages

Translation Strategy

TRANSLATION APPROACHES
=======================

Approach 1 — Professional Translation Agency:
  → Process: English article → Agency translates → QA reviewer approves → Published
  → Quality: Excellent (professional translators, domain expertise)
  → Speed: 2–5 days per article
  → Cost: $0.10–$0.25 per word ($30–$75 per article at 500 words)
  → Best for: P1 languages, high-visibility content, legal/compliance content
  → Volume: Top 200 articles × 3 languages × $50 avg = $30,000 initial investment

Approach 2 — AI-Assisted Translation (MTPE — Machine Translation, Post-Edited):
  → Process: AI translates → Human editor reviews and corrects → Published
  → Quality: Very Good (AI handles structure, human handles nuance)
  → Speed: 4–8 hours per article (50% faster than pure human)
  → Cost: $0.05–$0.15 per word ($15–$40 per article) — 40% savings vs agency
  → Tools: DeepL Pro, Google Translate API + human reviewer, Smartcat
  → Best for: High volume, P2 languages, ongoing content

Approach 3 — Community Translation:
  → Process: Native-speaking community members volunteer to translate
  → Quality: Variable (depends on contributor expertise)
  → Speed: Unpredictable (depends on volunteer availability)
  → Cost: Free or low-cost (recognition, small incentives)
  → Platform: Crowdin, Transifex, Lokalise (community translation features)
  → Best for: P3 languages, community-sourced content, supplementary articles

Approach 4 — Hybrid Model (recommended):
  → P1 languages (Spanish, French, German): Professional + AI-assisted
  → P2 languages (Japanese, Portuguese): AI-assisted + community review
  → P3 languages: AI-only with community flagging for quality issues
  → New articles: AI-first translation for all languages → human review queue
  → Cost optimization: $15–$40/article avg across all languages

TRANSLATION COST PROJECTION:
  ════════════════════════════════════════════════════════════════════════
  Language  | Articles | Cost/Article | Total Cost | Annual Update Cost
  ════════════════════════════════════════════════════════════════════════
  Spanish   | 200      | $35          | $7,000     | $2,100 (60 new/updated)
  French    | 200      | $35          | $7,000     | $2,100
  German    | 200      | $35          | $7,000     | $2,100
  Japanese  | 100      | $40          | $4,000     | $1,200
  Portuguese| 100      | $30          | $3,000     | $900
  Chinese   | 100      | $45          | $4,500     | $1,350
  ════════════════════════════════════════════════════════════════════════
  Total initial: $32,500
  Total annual maintenance: $9,750
  ════════════════════════════════════════════════════════════════════════

Translation Workflow

TRANSLATION WORKFLOW
=====================

New Article Translation Pipeline:
  1. English article published → Auto-flag for translation
  2. Priority assigned: Based on article category and search volume
  3. Translation triggered:
     → P1 languages: Professional translator assigned (within 1 business day)
     → P2 languages: AI translation generated → human reviewer assigned (within 3 days)
     → P3 languages: AI translation published (with "AI-translated" notice)
  4. Translation completed → QA review
  5. QA checklist:
     → Accuracy: Does translation match English meaning?
     → Completeness: All sections translated? (including screenshots)
     → Natural language: Reads naturally (not literal translation)?
     → Cultural appropriatenate: No awkward or offensive phrasing?
     → Technical terms: Correct domain terminology used?
     → Links: Internal links point to correct translated versions?
  6. Published to language-specific URL (/es/help/article, /fr/help/article)
  7. Search index updated for language-specific search

EXISTING ARTICLE UPDATE WORKFLOW:
  → English article updated → Translation status flagged as "stale"
  → "Last translated: [date]" shown to users (if translation is older than English)
  → Re-translation triggered within 30 days of English update
  → Urgent updates (breaking changes, security): Within 48 hours for P1 languages
  → Translation memory: Reuse previously translated segments (consistent terminology)

GLOSSARY AND TERMINOLOGY MANAGEMENT:
  ════════════════════════════════════════════════════════════════════════
  English Term      | Spanish     | French      | German       | Japanese
  ════════════════════════════════════════════════════════════════════
  Dashboard         | Panel       | Tableau de  | Dashboard    | ダッシュボード
                    | de control  | bord        |              |
  Two-factor auth   | Autenticaci| Authentifi- | Zwei-Faktor-| 二段認証
                    | ón de 2     | cation à    | Authentifiz  |
                    | factores    | cation 2    | iering (2FA) |
                    |             | facteurs    |              |
  Workspace         | Espacio de  | Espace de   | Arbeitsplatz | ワークスペース
                    | trabajo     | travail     |              |
  ════════════════════════════════════════════════════════════════════════
  → Maintain glossary in translation management system (TMS)
  → Enforce glossary terms in all translations (consistent across articles)
  → Review glossary quarterly; add new terms as product evolves

Multilingual KB Analytics

MULTILINGUAL PERFORMANCE METRICS
==================================

Usage by Language:
  ════════════════════════════════════════════════════════════════════════
  Language  | Sessions | Avg Pages | Search Success | Deflection | Ticket Creation
  ════════════════════════════════════════════════════════════════════════
  English   | 5,000    | 3.2       | 78%            | 42%        | 1,800/mo
  Spanish   | 1,500    | 2.8       | 72%            | 38%        | 600/mo
  French    | 1,000    | 2.9       | 74%            | 35%        | 400/mo
  German    | 800      | 3.0       | 76%            | 40%        | 300/mo
  Japanese  | 400      | 2.5       | 68%            | 30%        | 200/mo
  ════════════════════════════════════════════════════════════════════════

Content Gaps:
  ════════════════════════════════════════════════════════════════════════
  Metric                          | Value
  ════════════════════════════════════════════════════════════════════════
  Articles available in English   | 350
  Articles available in Spanish   | 280 (80% of English)
  Articles available in French    | 270 (77% of English)
  Articles available in German    | 260 (74% of English)
  Articles available in Japanese  | 120 (34% of English)
  Zero-result searches (Spanish)  | 12% (above target of < 10%)
  ════════════════════════════════════════════════════════════════════════
  Action: Translate 70 more Spanish articles (gaps driving zero-results)

ROI by Language:
  → Spanish KB: 1,500 sessions/month × 38% deflection = 570 deflected tickets
     Deflection savings: 570 × $8.50 = $4,845/month = $58,140/year
     Translation cost: $7,000 initial + $2,100/year
     ROI: 764% annual

  → Total multilingual KB savings: ~$180,000/year across all languages
     vs Total translation cost: ~$42,250 (first year)
     Net first-year savings: $137,750

Integration Points

Edge Cases

Disclaimer: All rights reserved by Circulos AI. These skills are specifically designed for Claude Code, Claude Cowork, Codex, and OpenClaw. When using or referencing any skill, please provide proper attribution to Circulos AI.