A plain-English guide for the ops team.
This tool helps you figure out how much dry ice to order each week for each fulfillment center. It replaces the manual spreadsheet process with an automated system that watches the same things you do — order volume, weather, transit times, partner estimates — and gives you a specific recommendation with its reasoning.
The goal is simple: order enough ice so nothing ships warm, but don't order so much that you're paying for ice that just evaporates.
You're still in control. The system makes recommendations, but you approve (or adjust) every order. Think of it like a co-pilot that does the math and watches the forecast while you make the final call.
The dry ice ordering cycle follows the same pattern you already use:
Main ordering day. The system generates a forecast report for next week (Mon–Sat) that starts with projected weekend carryover — Friday's ending balance minus ~25% weekend sublimation loss. You review the recommended amounts for each FC, adjust if needed, and place your orders.
Mid-week check-in. The system compares what actually happened Mon–Wed against the forecast. If you're burning through ice faster than expected (heat wave, promo spike), it'll recommend a Monday top-up order.
The dashboard updates every day showing ice on hand, ice used, and what's predicted for the rest of the week. Alerts pop up if any FC is projected to run short.
The system combines several data sources to predict how much ice each FC needs:
System Most of next week's orders are subscription renewals that are already scheduled in the system. On a typical week, these make up 70–80% of total volume. The system counts all orders with a fulfillment date in the target week, broken down by FC and day.
AI Not all orders are subscriptions — some are one-time purchases from the website. The AI looks at the last 4–6 weeks of order patterns to estimate how many new orders will come in each day. It also accounts for the fact that some subscription orders will get rescheduled or cancelled before they ship (historically about 8–12% depending on the week).
You For FCs that handle partner orders (like BYB at Dallas), you enter the expected order counts that partners share with you. You can type notes in plain English ("BYB expects 200 orders, heavier toward the end of the week because of a promo") and the AI will extract the numbers and factor them in.
System This uses the same ice charts we've always used. For each order, the system looks at the destination weather forecast and transit time to determine how many pounds of dry ice to pack. Hotter weather = more ice. Longer transit = more ice. The system knows the typical destination mix for each FC (e.g., Dallas ships mostly to the South and Southwest where it's warmer) and uses weather forecasts for those regions.
System Dry ice evaporates — about 12–15% of its weight each day just sitting in the freezer. The system tracks this daily loss so the "ice on hand" number stays accurate between physical counts.
Weekend carryover is critical: The system takes Friday's ending ice balance at each FC, applies ~25% sublimation loss over the weekend (Sat–Sun), and calculates Monday's starting balance. This is the foundation for next week's forecast. For example, if Dallas ends Friday with 900 lbs, the system projects ~680 lbs on hand Monday morning. A large carryover means you order less next week; a near-zero carryover means you need a bigger order. The forecast report shows this calculation explicitly for each FC.
AI The system adds it all up: ice needed for predicted orders, minus ice on hand, minus ice already ordered, plus a buffer for sublimation and uncertainty. It gives you a specific number for each FC with an explanation of how it got there and a confidence level (high, medium, or low depending on how much of the forecast is based on confirmed orders vs. estimates).
The system handles the math, but it needs a few inputs from you to stay accurate:
| Task | When | Why It Matters |
|---|---|---|
| Enter physical ice counts | Monday and Friday (or whenever you count) | The system calculates what it thinks is on hand, but only a real count keeps it honest. If the system says 3,200 lbs but you count 2,900, the forecast adjusts automatically. |
| Enter 3PL partner estimates | As soon as you hear from them (usually Tue–Wed for next week) | Partner order volumes can't be predicted from our data. When BYB says "expect 200 orders next week," entering that note is what makes the forecast accurate for Dallas. |
| Record ice purchases | When you place an order with the vendor | The system needs to know what ice is arriving and when, so it can project inventory forward. |
| Review the Thursday forecast | Thursday morning before ordering | You make the final call. If something looks off (you know about a promo the system doesn't, or a vendor issue), adjust the amounts before approving. |
| Add notes for anything unusual | Anytime | Big promos, holidays, vendor delivery changes, anything the system can't see on its own. Write it in plain English — the AI reads these notes when building the forecast. |
Every day, the system records what it predicted vs. what actually happened. This creates a track record:
The "Forecast Accuracy" section on the dashboard shows last week's predictions vs. actuals, so you can see how well the system is doing and whether it's improving.