Hello! I'm Keita, a Backend Engineer at Flagship Inc.

With AI-related research winning both the Nobel Prize in Physics and Chemistry this year, I'm reminded once again of AI's tremendous impact across industries.

At our company, AI adoption continues to advance steadily—we've built internal chatbots that can answer company policy questions and regularly utilize AI assistance in code development.

Today, I'd like to share a case study from one of our recent Shopify migration projects where we leveraged AI to achieve significant operational efficiency improvements for a client (referred to as Company A).

Mass Translation of Product Detail Information

Company A operates globally with over 10,000 products in their e-commerce catalog. Global expansion means customers from various countries visit their online store, requiring the site to display in their respective local languages.

Visitors from Spain expect Spanish, French customers expect French, and American customers expect English interfaces.

Each of the 10,000+ registered products includes detailed descriptions covering materials, features, and selling points—all designed to help customers understand product value and appeal.

Displaying 10,000+ products across multiple languages presents an obviously complex challenge.

Our initial approach involved standard automated translation services, but these failed to meet Company A's quality requirements.

The issue was that product descriptions contained specialized terminology that automated translation services handled poorly, producing awkward or incorrect translations.

Publishing such poor translations could undermine customer trust and brand credibility.

ChatGPT Implementation

This led us to implement ChatGPT as our solution!

We provided ChatGPT with specific instructions for handling specialized terminology translations. Additionally, since translated content needed to include summaries and bullet-point formatting, we incorporated these requirements into our prompt instructions.

The process became straightforward: copy product descriptions, paste them into ChatGPT, and request translation according to our specified guidelines.

Challenge #1: Scale and Timeline Constraints

We encountered our first major obstacle: translation volume exceeded manual processing capabilities. Individual manual ChatGPT translations would clearly miss our project deadline. This challenge led us to develop automated translation using ChatGPT's API.

Google Sheets + GAS + ChatGPT API (Chat Completions API)

Are you familiar with GAS (Google Apps Script)?

It's a powerful automation tool for Google services including Sheets, Gmail, and Calendar!

Common applications include:

1. Automatically emailing information entered in spreadsheets
2. Weekly automated calendar summary emails
3. Automatic "thank you" emails when forms are submitted

Using GAS, we developed a system that automatically translates product descriptions entered in spreadsheet cells via ChatGPT's API.

We implemented the Chat Completions API, which provides programmatic access to the ChatGPT conversational interface we use daily.

Our GAS development referenced this helpful article.

The system automatically processes product descriptions entered in cell A1, applying our ChatGPT instructions to output translations in cell A2.

This automation enabled us to handle 10,000+ translations efficiently. Problem solved—or so we thought.

Unfortunately, we encountered another obstacle.

Challenge #2: API and Service Limitations

Our solution hit limitations in both GAS and ChatGPT API constraints, causing processing errors.

Specifically, GAS URL Fetch has a maximum 50MB request size limit for Workspace accounts, with concurrent execution restrictions.

OpenAI's Chat Completions API also imposes limits of 200 requests per minute or 40,000 tokens per minute.

Batch API Solution

To circumvent these limitations, we implemented OpenAI's Batch API, which offers:

• 50% cost reduction compared to standard API pricing
• Separate, higher rate limits
• Asynchronous processing completed within 24 hours

This approach perfectly suits scenarios where immediate responses aren't required, but cost-effective mass data processing is essential.

Exactly our use case.

Our architect and GAS specialist Reona implemented this solution, successfully enabling mass data processing.

Thank you for your invaluable contribution despite your busy schedule.

Technical Summary and Insights

This case study demonstrates how we leveraged AI to efficiently translate 10,000+ product descriptions across multiple languages.

Through iterative problem-solving, we ultimately succeeded by combining GAS with OpenAI's Batch API. A key insight: Batch API proves highly effective for mass ChatGPT data processing requirements.

If you encounter similar challenges, consider implementing this approach. We'd love to hear about alternative solutions—please contact us with your experiences.

Until next time!