The Problem Starts: Buyer Orders Were Always 「Off」
Every Monday morning, here it comes. Opening the inbox, the numbers never add up. Buyer A writes 「100」 in the quantity field, but says 「1000」 in the email body. Buyer B orders the same product twice within a day. Intentional or a mistake.
At first, we checked each one manually and then reached out. Average 3 to 4 error reports daily. Over 100 per month.
「Is there a way to catch these automatically?」
Before: The Manual Review Trap
• Buyer sends order via email
• Team manually inputs into spreadsheet
• Errors discovered 3 to 4 hours later
• Email sent requesting corrections
• Wait for buyer response (usually 24 hours)
• Re-enter data
Every cycle delayed shipping preparation.
Solution: Google Forms + Claude Code Combo
Step 1: Plant 「Validation Rules」 in Google Forms
In Google Forms advanced settings, we set conditions for each field.
Quantity field → Numbers only, range 1 to 9999
Email field → Email format validation
Product code → Only registered products (dropdown)
Delivery date → Only dates 7+ days from today
When buyers tried to break the rules, they'd instantly see 「Please re-enter.」
Step 2: Claude Code Performs 「Context Validation」 When Responses Arrive
Errors that Google Forms can't catch (duplicate orders from same buyer, impossible product combos) are Claude Code's job.
The automation agent on the Windows environment reads the Google Sheet every 5 minutes:
• Detects duplicate orders from same buyer
• Finds orders nearly identical to last week's
• Catches orders when inventory is low
• Spots impossible delivery requirements
When it finds something, instant Slack message and auto-reply email:
Subject: [⚠️ Order Review] Buyer Name
Body: "Quantity" is outside normal range.
This is XX times more than last week's order.
Please confirm.
Step 3: Buyer Responds Instantly
When buyers get the auto-message, many reply with corrections within 30 minutes.
Manual inquiries used to take 18 hours average.
After: Automated Review Process
| Metric | Before | Now |
|--------|--------|-----|
| Time to detect error | 3~4 hours | 5~10 minutes |
| Buyer confirmation time | 18 hours | 30 minutes |
| Errors team handles monthly | 100+ | 8~12 |
| Shipping delays (error-caused) | 2~3/month | Nearly zero |
Result: Freed up 5 hours per week from "error email ping-pong."
The Unexpected Discovery
One thing got interesting.
As Claude Code logged errors, it started finding 「patterns.」 Some buyers always rush orders Monday morning. Others split the same product into multiple orders.
When we looked at this data and talked to buyers, there were reasons.
• Buyer A: 「Our weekly plans change constantly.」
• Buyer B: 「One big order feels risky, so we split it.」
Automation didn't just reduce errors. It deepened our conversation with buyers.
The Bottom Line
Validation that started with one Google Form, when it met Claude Code as a small helper, redrew the line between 「what humans should do」 and 「what machines should do.」
Errors are no longer our mistakes. They're clues we solve together.