By September it is already too late. The order was due months ago.
# units you actually sell per day ADD = units_sold / days_in_window # windows that match the SKU evergreen 30-60 days trending 14 days seasonal same period last year
# reorder point ROP = (ADD × lead_time) + safety_stock # worked: 25/day, 75-day lead, # 14-day buffer ROP = (25 × 75) + 350 = 2,225 units
order_qty = ADD × (lead_time + reorder_cycle + safety_days) − (on_hand + inbound) # safety stock buffers spikes in # BOTH demand and lead time
| Date (2026) | Event | Why it matters |
|---|---|---|
| ~Oct 6-14 | Prime Big Deal Days | The October "second Prime Day" (est.) |
| ~Oct 19-23 | Amazon inbound cutoff | Must ARRIVE by this date (2026 est.) |
| Nov 27 | Black Friday | Early Thanksgiving = compressed runway |
| Nov 30 | Cyber Monday | Peak of peak |
Note: the summer Prime Day was June 23-26, not July. Amazon's exact 2026 cutoff is not published yet.
| Production | 4-6 weeks |
| Ocean (West Coast) | 15-25 days |
| Drayage + prep | 3-7 days |
| Amazon check-in | 6-7d median, tail to 30+ |
Honest version: hybrid beats pure AI. The model computes and drafts. You approve.
You, in Slack: "Skip the 1-unit product, order 50 each of the other two, call it 'Slack Demo', export a PDF, draft the email."
Four systems, one chat: inventory, PO, document, email. The agent acts. You approve.
Let InventoryHero watch the loop and hand you the POs.