Why We Kept Checking Manually
Every Monday morning was chaos. We opened Google Sheets and manually calculated last week's sales volume, remaining warehouse inventory, and reorder timing. With over 100 product types, it was easy to miss something. Once, we didn't notice a critical item was almost out of stock until Thursday.
「Could we automate this repetitive work」
Before: The Limits of Manual Checking
• Manual Google Sheets review every Monday (about 60 minutes)
• Comparing sales and inventory visually, then consolidating data in spreadsheets
• Irregular reorder timing (prediction failure rate around 40%)
• Missing at-risk items (average 3 to 4 days late detection)
After: Automated Alert System
Google Sheets (input sales data)
↓
Claude Code (automatic analysis)
↓
Auto-assign color tags
├─ 🟢 Normal (sufficient stock)
├─ 🟡 Caution (reorder within 2 weeks)
└─ 🔴 Critical (immediate reorder)
↓
Right column auto-calculates recommended order qty for next week
What the Bot Learned
The first week was just simple quantity comparison. But watching the data patterns over time revealed something interesting about our business.
• **Seasonal trends**: Higher sales Tuesday through Thursday
• **Product rotation rates**: Some items sell 30 units per week, others only 5
• **Shipping lead time**: Average 10 days from reorder to delivery (so we must order 10 days ahead)
By week two, the bot stopped just comparing numbers and started automatically calculating "when will this item run out."
Accuracy Improvement Timeline
| Week | Prediction Accuracy | Order Failures | Overstock |
|------|-------------------|----------------|----------|
| 1 | 45% | 5 cases | 3 cases |
| 2 | 62% | 2 cases | 2 cases |
| 3 | 71% | 1 case | 1 case |
| 4 | 78% | 0 cases | 0 cases |
Unexpected Side Effects
1. **Stock status is always visible**: Our email automation bot also reads the sheet and instantly knows if a product is available to quote to buyers
2. **Order records auto-save**: After 4 weeks, we can instantly look up "what did we order when and how much"
3. **Team communication dropped**: Inventory questions in our chat nearly disappeared
Still Needs Work
• Sudden large orders can't be predicted yet (happens 1 to 2 times a year)
• Returns and damaged items still require manual adjustment
• Auto-calculating reorder amounts when exchange rates fluctuate isn't smooth
Next Experiment
We're testing whether a Mac mini can read this sheet data and automatically sync it with our Windows device's ordering system. If it works, will purchase orders generate automatically the moment we input data into the sheet