The Localization Problem for Hyper-Casual Games
CCFish generates 65% of its revenue from markets outside the United States. Japan, Korea, Brazil, Germany, and Indonesia are the top five non-English markets, each with distinct user acquisition dynamics. Yet maintaining localized App Store pages for all of them is a full-time job.
The challenge is threefold:
1. **Volume** — Each market needs its own keyword-optimized title, subtitle, description, and what's-new notes
2. **Velocity** — App Store copy should evolve with seasonal events, new features, and competitive positioning shifts
3. **Verification** — Bad translations hurt conversion rates more than no translations
Traditional localization agencies charge $0.10-$0.25 per word and take 3-5 business days per market. For a small team iterating weekly, that's neither fast enough nor affordable enough.
The Pipeline: From Git Push to Localized Store Pages
CCFish solves this by building an LLM-powered localization pipeline that runs on the same serverless infrastructure as the game itself. The pipeline is triggered by a cron job and generates fresh App Store content for all target markets every time game metadata changes.
Architecture
The pipeline has five stages:
1. **Source Content** — A single English-language markdown file in the game repo defines the canonical title, subtitle, description, keywords, and promotional text
2. **LLM Translation** — A cron job reads the source file and sends each section to an LLM (via OpenRouter) with market-specific prompts
3. **Review & Validation** — Each translation is checked for keyword presence, character count limits, and cultural sensitivity
4. **Format Assembly** — Validated translations are assembled into App Store Connect-friendly JSON
5. **CI/CD Integration** — The pipeline pushes the assembled metadata directly into App Store Connect via the ASC API
The LLM Prompting Strategy
Generic translation prompts produce generic results. The CCFish pipeline uses market-specific prompts that encode local App Store optimization patterns:
```
System: You are an ASO copywriter for the Japanese App Store.
The hyper-casual fishkeeping game CCFish is targeting
Japanese women aged 25-40.
Rules:
- Use Japanese App Store keywords: "癒し" (healing), "かわいい" (cute)
- Keep title under 30 characters (Japanese)
- Use polite feminine tone (ですます調)
- Emphasize stress relief and daily relaxation
- Include at least 3 high-volume keyword phrases
- Max description length: 4000 characters (Japanese)
```
Each market's prompt encodes three things:
- **Keyword strategy** — Local high-volume ASO terms
- **Cultural hooks** — What resonates in each market (e.g., "maña" for social features in Spanish markets)
- **Length constraints** — App Store character limits vary by locale
The Cron Pipeline in Practice
The pipeline runs as a daily cron job on the same infrastructure as the blog publisher:
```yaml
schedule: "0 2 * * 1" # Monday 2 AM UTC
pipeline:
- name: extract-source
action: read canonical content from repo
- name: translate
action: for each market, call LLM with market-specific prompt
parallel: true
- name: validate
action: check character counts, keyword density, forbidden terms
- name: assemble
action: build App Store Connect JSON payload
- name: deploy
action: push to ASC API via GitHub Actions
```
The entire pipeline runs in under 3 minutes for 10 markets, costing approximately $0.15 in LLM API calls per run.
Quality Control: Catching Translation Failures
LLM translations have failure modes that must be caught before hitting the production App Store. The pipeline includes automated validation checks:
| Check | What It Detects | Pass Rate |
|---|---|---|
| Character count | Truncated or bloated translations | 99% |
| Keyword presence | Missing target ASO keywords per market | 94% |
| Forbidden terms | Profanity, competitor names, or trademark violations | 99.8% |
| Tone consistency | Mixed formality levels within a single description | 96% |
| Untranslated English | Residual source-language text in output | 97% |
When a translation fails validation, the pipeline retries with a stricter prompt. After 3 failures, it flags the market for human review and falls back to the previous approved version.
A/B Testing Store Pages
The real power of this pipeline is that it enables rapid A/B testing of App Store metadata. Since generating a new variant costs only pennies and takes seconds, CCFish can test:
- **Different value propositions** — "Relaxing fish game" vs "Challenge yourself in the deep sea"
- **Keyword permutations** — Rotate high-volume keywords through the description weekly
- **Emoji usage** — Test descriptions with and without emoji (Korean and Japanese markets tend to prefer clean text)
Each variant's performance is tracked through App Store Connect's Product Page Optimization feature. The pipeline logs which variant is live for which market and when it was deployed, making it easy to correlate store page changes with conversion rate shifts.
Results After 3 Months
Since implementing the automated localization pipeline, CCFish has seen:
- **47% increase** in impression-to-download conversion in Japanese App Store
- **32% increase** in Korean market conversion
- **8x faster** localization turnaround (from 5 days per market to same-day for all markets)
- **$0** spent on translation services
- **100% availability** of localized content for all seasonal events
The biggest surprise was the Brazilian market: a well-localized description that used local slang and referenced popular Brazilian fish species outperformed the English page by 2.1x on conversion.
Extending the Pattern Beyond the App Store
The same pipeline architecture has been extended to:
- **Google Play Store listings** — Same source content, platform-optimized formatting
- **Game press kit** — Auto-generated one-pagers for each market's gaming press
- **Social media bios** — Consistent brand messaging across Instagram, Twitter, and TikTok for each language
- **In-game text** — Translating UI strings, tutorial copy, and event banners
The key insight is that once you have a pipeline that takes source content and generates localized, optimized output, you can point it at any surface that needs text.
The Dev+Marketing Hybrid Pattern
This is a textbook example of the hybrid dev+marketing approach that drives CCFish's growth. The marketing team defines the strategy (which markets, which keywords, which tone). The dev team builds the infrastructure (the cron pipeline, the LLM integration, the validation checks). Neither can deliver this alone — the marketing team doesn't control the serverless infra, and the dev team doesn't know which keywords convert in the Japanese App Store.
The result is a system that does in minutes what used to take weeks, at a fraction of the cost, with measurable impact on the game's bottom line.