The Problem Started With One Small "Blank Field"
We receive over 100 Google Form submissions every week. The「Special Requests」field that each buyer leaves usually contains just 2 to 3 lines, and we spend time reading through to figure out what we still need to ask.
For example, one buyer wrote only「Red shirts, mixed sizes」and we had to manually think through:
• What shade of red (bright or deep?)
• Size ratio (how many of each size?)
• Shipping deadline (by when?)
• Do we need to confirm minimum order quantity?
This process repeated constantly, leading to errors and delayed responses.
The Solution: Claude Code Handling the Middle Step「Read, Judge, Generate Questions」
We set up Google Form responses to automatically trigger Claude API calls through Apps Script.
Claude's instruction text:
「Read this buyer's message and generate 3 questions we might have missed.
Format: Question (English) | Suggested answer options」
Results appear directly in Google Sheets, and we simply copy those questions into our follow-up email template.
Before / After
Before (Manual Review)
• Review time per form response: 5-10 minutes
• Weekly processing time: 8-10 hours
• Follow-up inquiries due to missed questions: 3-5 per week
• Average buyer response wait time: 2-3 days
After (Claude Code Auto-Review)
• Review time per form response: 1-2 minutes (Claude already suggested answers)
• Weekly processing time: 1.5-2 hours
• Missed questions: 0.5-1 per week (almost none)
• Average buyer response wait time: Within 2 hours (we send immediately)
Unexpected Benefits
1. **Consistency Improved**: Claude reads all buyer messages by the same standard, so nothing falls through the cracks due to individual oversight.
2. **Buyer Satisfaction**: Receiving precise follow-up questions quickly, buyers now see us as「thorough sellers」.
3. **Team Capacity**: Those 8 hours freed up for other tasks (email drafting, sample preparation), so overall pace increased dramatically.
One Small Failure: We Generated "Too Many Questions" At First
Initially I told Claude「generate all necessary questions,」and it asked one buyer 10 questions. The buyer was shocked (laughs). After that, we changed it to「between 3 and 5,」and asked Claude to prioritize them.
The Key Takeaway
Non-developers can use Claude Code (or Apps Script + Claude API) to automate those「middle steps that require human judgment」. It's not complete automation, but when we pre-sort our thinking process before starting, both speed and accuracy improve dramatically.
Next, we're testing automated categorization of buyer responses.