Supply Chain / Operations
Case 06 / 10
Demand planning and reorder logic
Twelve-month rolling forecasts connecting commercial demand to inventory, purchasing, and field capacity.
Crave Robotics · 2026 · Sole owner
- 12-month rolling forecasts wired directly into purchasing.
- Reorder timing driven by projected stockout dates, not a fixed calendar.
- Zero stockout-driven lost sales through a 67% fleet expansion.
The numbers
- 12 months rollingForecast horizon
- Stockout-projection drivenReorder logic
- 0Stockout-driven lost sales in period
- TypeScript
- Node.js
- PostgreSQL
- Scheduled jobs
Explore the platform
Interactive — synthetic dataClick through the sidebar — each module is a real screen from the platform, populated with synthetic data.
12-Month Plan
| SKU | Current velocity | 12-mo forecast | |
|---|---|---|---|
| Garlic Chicken Pasta | 62/wk | 88/wk | |
| Beef Burrito | 40/wk | 35/wk | |
| Veggie Bowl | 51/wk | 70/wk | |
| Garlic Noodles | 28/wk | 24/wk |
This is the actual Demand Planning module of the operating platform, opened to the 12-Month Plan screen. SKU names and velocity numbers are synthetic; the screen and its layout are real.
How it fits together
- 1Telemetry
- 2Per-product velocity
- 3Seasonality-adjusted projection
- 4Stockout date per SKU
- 5Reorder trigger against lead time
- 6Purchasing recommendation
Read the full case study Hide the case study
The problem
Purchasing and inventory decisions for a growing fleet had no forward-looking demand signal. Reorder timing was reactive, and nothing connected planned machine installations to the inventory they would require once live.
Why it mattered
Frozen food carries expiration risk on one side and stockout risk on the other, with vendor lead times in between. Without a forecast the business was choosing between waste and lost sales, and installation plans were being made without checking whether the inventory to stock them existed.
What I built
Twelve-month rolling demand forecasts using per-machine per-product velocity with seasonality, extended into the decisions that actually consume a forecast:
- Velocity-based out-of-stock projection per SKU per machine
- Reorder cadence driven by projected stockout timing rather than a fixed schedule
- Installation-schedule integration, so a machine going live next month carries demand this month
- Exception flagging when inputs or outputs look wrong
- Product-level planning parameters — vendor lead times, shelf life
The shift from fixed cadence to projection-driven ordering is the part that mattered. A fixed schedule orders the same thing every cycle regardless of what is actually selling; a projection orders what will run out, when it will run out, offset by how long the vendor takes.
What I learned
The forecast was only useful once it was wired to a decision. Projecting demand was straightforward; the value came from converting a projection into a dated purchase order recommendation against a specific vendor lead time.
What I would do next
Backtesting infrastructure to measure accuracy by horizon. Per-SKU method selection for intermittent demand. Explicit safety-stock optimization against service-level targets rather than threshold heuristics.
Described without confidential information. The demo and any figures inside it use synthetic data.