30 Minutes Every Morning , Until We Finally Spotted the Problem
Every morning, a single spreadsheet appeared on the Windows device monitor. New order information arrived mixed together. Our small team would open each item one by one, judge its status, then move it to the corresponding folder. Each batch typically took over 30 minutes.
「Can we automate this?」
At first, it felt like a complex problem. But after talking with Claude Code, we found the real issue.
The problem wasn't a lack of rules. The rules were just stuck inside our heads.
Making Rules 「Visible」 in the Spreadsheet
Our first step was simple. We added a new column to Google Sheets and explicitly wrote down the filtering criteria.
Like this:
Order ID | Status | Region | Priority | Bot Decision
12345 | Pending | Asia | High | ✓ Process Now
12346 | Confirmed | Europe | Medium | ○ Wait (Review)
12347 | Rejected | Americas | Low | ✗ Hold
When we showed these rules to Claude Code, the bot started recognizing patterns. Based on combinations of status, region, and priority, it began automatically deciding the next steps.
The 3-Second Miracle
When we saw the first automated result, there was only one thing to say:
「Really?」
The entire spreadsheet was classified in one go. No need to click through each item anymore. Google Sheets filtering became the bot's eyes and its decision-making criteria.
Here's what we measured:
• **Manual classification**: ~30 minutes
• **Claude Code automatic classification**: ~3 seconds
• **Accuracy**: Initial 92%, after adding rules: 98%
Why Non-Developers Can Do This
What makes this system special is that no programming knowledge is required.
You only need two things:
1. **Can you write the rules on paper?**
- What criteria do I use when doing this work myself?
2. **Can you transfer those rules to the spreadsheet?**
- Add columns, input a few examples, then show Claude Code
Claude Code reads your written rules and automatically applies that same logic to all the remaining data.
What We Learned From Failure
Our first attempt failed. We put in too many conditions at once, and the result was chaos. Even the bot got confused.
So we simplified the rules:
• 5 conditions → reduced to 3 core conditions
• Exception cases moved to a separate sheet
• Review misclassified items once per week
After this, accuracy skyrocketed.
What We Do Now
This automation system runs every single day. One of our dream team bots checks the spreadsheet at dawn, automatically classifying according to rules.
What we do in the morning is simple. Open the sheet, check at a glance if everything classified correctly, then manually fix only the errors.
In the end, automation isn't about programming. It's about organizing rules.