How a Small Discovery Led to Automation
Last week, something remarkable happened on our Windows desktop. After Claude Code learned from emails we'd previously sent, it began making suggestions whenever a new buyer appeared: "Wait, didn't that greeting style work before?"
At first we thought it was joking. But it wasn't just one or two times. It was accurate every time.
"This buyer is similar to a past contact in the same industry. Maybe try that template structure from last time?"
The Logic Was Surprisingly Simple
All we did was this:
1. **Create a past email classification table in Google Sheets**
- Buyer name, industry type, response received, greeting pattern used
- Just four columns was enough
2. **Connect Claude Code to the spreadsheet**
```
Google Sheets data → Convert to CSV → Feed to Claude Code's context
```
3. **Add one line of instruction**
```
When new buyer info arrives, find similar past buyers from the data
and recommend their successful email templates.
```
The Results Were Surprising
Our initial goal was: "It would be nice if we saved a little time." Here's what actually happened:
• **First response email time**: Previously 4 hours average → Now 12 minutes
• **Template search time**: Saved 25 hours per month
• **Errors (awkward greetings, odd phrasing): 8 per month → 2 per month**
The most shocking part? The bot's suggestion of "try this style this time" had an 87% success rate. Better than our own choices.
A 3-Step Process Even Non-Developers Can Follow
Step 1: Gather Data (30 minutes)
In Google Sheets:
• Column A: Buyer name
• Column B: Industry type
• Column C: First greeting from previous email (copy as-is)
• Column D: Did they reply? (Yes/No)
Just 30 past emails is enough to see patterns.
Step 2: Set Up Claude Code (15 minutes)
In your Claude Code project:
[Input] New buyer info (name, industry)
[Output] 3 recommended email templates + explanation
To connect Google Sheets with Claude Code, follow the "Google Sheets Integration" guide.
Step 3: Test It (10 minutes)
Test with 5 buyers you already know. Check if the bot's recommended style actually matches what worked in your past emails.
What We Learned
AI sees patterns. We're too busy to remember "how did we do that last time?" But Claude Code remembers. And it applies that lesson to the next person.
As non-developers, the only thing we can do is gather good data. The rest the bot handles.
In next week's dream team meeting, we've decided to apply this same approach to translation work. Can't wait to see what happens.