The Problem: A Sea of Order Notes
Every morning, the Dream Team faced dozens of new orders. But the real headache was order notes. Buyer messages were all over the place with no consistent format.
「Color change needed」
「URGENT, need confirmation by tomorrow」
「Attached sample, please reference」
「Same as previous order」
The same type of request appeared in different ways, and priority levels were mixed throughout. Spreadsheet filters and formulas had clear limitations. When the thought "this is a waste of time" crossed our minds, we met Claude Code.
Bot Design: From Text Understanding to Priority Scoring
The core concept was straightforward.
1. Read order note text
2. Auto-detect the note type (color change / urgent / sample reference / other)
3. Assign a priority score
4. Automatically record to Google Sheets
Input: 「Need color adjustment and quick sample production by tomorrow」
↓ Claude Code Analysis
→ Classification: Urgent + Color Change
→ Priority Score: 9/10 (high urgency)
→ Auto-entered in Google Sheets
Implementation: A Non-Developer's 4-Day Journey
Day 1: Google Sheets Setup (30 minutes)
We created a new spreadsheet and added these columns.
• Order Number
• Original Note
• Auto Classification (result cell)
• Priority Score
• Assigned Owner (manual)
Day 2: Writing Claude Code Prompts (2 hours)
We refined the prompt in Claude Code. The key was instructing it to treat the notes as messages from buyers at a trading company, with clear classification criteria.
Classification Categories:
• Color/Size Change Requests
• Urgent Processing (tomorrow, this week, etc.)
• Sample/Test Requests
• Reference Existing Order
• Other
Priority Scoring (1~10):
+ 3 for urgent expressions
+ 2 for color/size changes
+ 1 for sample requests
Day 3: Google Sheets API Integration (1 hour)
We connected Google Sheets API to Claude Code. As a non-developer, Claude auto-generated the necessary code snippets.
# Code snippet generated by Claude (no modifications needed, just copy-paste)
from google.colab import auth
auth.authenticate_user()
import gspread
gc = gspread.authorize(creds)
ws = gc.open("Order Note Classification").sheet1
# Read each row's note and analyze
for row in ws.get_all_records():
memo = row['Original Note']
# Execute classification logic
classification = analyze_memo(memo)
ws.update_cell(row_idx, col_idx, classification)
Day 4: Execution and Refinement (1 hour)
The first run achieved 90% classification accuracy. Expressions like "very urgent" weren't caught. We added a synonym list to the prompt, bringing accuracy up to 96%.
Real Results: Time and Stress Reduction
Before Implementation
• 40 minutes daily spent classifying order notes
• Human errors caused important requests to be missed
• Team members reviewed the same notes multiple times
After Implementation
• Classification time reduced to 5 minutes (fully automated)
• Consistent priority standards applied
• Team members only review pre-classified notes
• Monthly "response time" cut by 30%
What a Non-Developer Learned
This journey taught us several key lessons.
1. **Text analysis is about context, not rules.** Spreadsheet functions only find exact matches, while Claude understands meaning.
2. **Prompts improve iteratively.** No need to be perfect on day one. After a few runs, patterns like "this one gets missed" become obvious.
3. **Real automation value lies in consistency.** Time savings matter, but processing all orders by the same standard was the bigger win.
What's Next
Currently the bot only classifies notes, but we plan these expansions.
• Auto-assign owners based on priority level
• Send urgent orders to the team via notification bot
• Identify recurring request patterns and suggest auto-reply templates
The "built in 4 days" headline is accurate, but our planning drew on 3 months of prior experience. That foundation made the speed possible.
Stay tuned for the Dream Team's next automation challenge.