Problem: 500 Emails, 5 Minutes of Manual Classification Daily
A B2B trading team receives dozens of buyer inquiry emails each week. Each message requires judgment about its status.
• Are we still waiting for a reply
• Is the deal confirmed
• Or is it on hold
Until now, the person in charge manually entered the status next to each subject line. Reading complex email content and making judgment calls took considerable time.
Morning email sorting took 5 minutes, afternoon sorting another 5 minutes. Almost an hour per week.
Solution: Claude Code Bot's "Read and Classify"
When buyer email content is copied into Google Sheets, the Claude Code agent automatically:
1. Reads the email body
2. Determines the final action (question pending, awaiting confirmation, deal finalized, etc.)
3. Auto-fills the status column ("Awaiting Reply", "Deal Confirmed", "On Hold")
There were some errors at first. When emails came in Chinese, the bot sometimes misunderstood the status; emails about renegotiating terms got marked as complete. But after adjusting instructions every 5 to 10 examples, accuracy improved significantly.
Results: Zero Manual Work, Higher Quality
Now, every Monday morning, the bot runs a batch classification. The team member only spot-checks unusual items (usually 2 or 3).
Before:
• Manual classification: 1 hour per week
• Error rate: roughly 5% (1-2 errors per 25)
After:
• Auto-classification: 0 minutes (batch process)
• Review time: 2 minutes per week
• Error rate: under 1%
Small Insight: Culture and Email Tone
What became interesting is that different buyers have different email styles. Some are direct ("Your price is too high"), others are indirect. Claude Code took a few days to learn these differences.
From a non-developer's perspective, building a bot isn't about creating it from scratch, it's about "watching what the bot misses." That turned out to be more engaging.
Next Steps
Right now we're only doing status classification, but next we want to auto-generate reply drafts for each status. "Awaiting Reply" emails could get a template like "We'll respond by next Thursday", while "On Hold" gets "We're currently reviewing this."
Small automations add up. Hidden time emerges for the team.