The Problem: "Text Noise"
Incoming order notes arrived in wildly different formats. "Fast shipping", "Extra durable product", "Can you match the color?", "What's the lead time?", "Multiple items in one order" , all mixed together. Sorting them manually took 20 minutes every single day.
What We Tried: A Simple Claude Code Prompt
We set up a Claude Code agent to run whenever a new order appeared in Google Sheets:
Read the order notes field and classify it into one of these: Shipping-related, Product Customization, Price Inquiry, Quantity Inquiry, Other
Google Sheets → Claude Code Agent
Input: A2 (order note text)
Process: Use prompt to determine category
Output: B2 (auto-fill classification result)
The Surprising Moment
After one week, Claude Code started to "learn" patterns. Initially it caught only obvious keywords ("fast", "color", "price"). Later, it even recognized slang like "doesn't wrinkle easily" and correctly labeled it as "Shipping-related".
Week 2: Accuracy exceeded 95%. Only about 5 entries needed manual fixes.
Why This Matters
• Time saved: 20 minutes → 3 minutes
• Fewer human errors: No more inconsistent classifications of the same request type
• Usable data: Instantly see which request types are most common
What's Next (Currently Testing)
We're linking the most frequent categories to pre-built reply templates and auto-suggesting response messages. Early results look promising.
Small Lessons Learned
Claude Code's real strength is "no need to hardcode rules". Feed it a few example rows in Google Sheets, and it learns from context. It's not perfect, but for non-developers, it's more than enough.
Start small with a prototype, feed it real data, and see how it responds. That seems to be the secret of non-developer automation.