Doudlah operates a complex, multi-stage agricultural supply chain where the product moves through several distinct physical states before reaching the end customer. Because planting decisions must be made months before sales occur, the supply chain is fundamentally asynchronous — market demand and production inputs are decoupled in time. Mark is developing a predictive inventory model to bring intelligence to this chain, starting with the Amazon fulfillment channel.
See also: [1] | [2]
The full production flow from field to customer:
Farming → Raw Product → Cleaning → Clean Product → Packaging → Packaged Product → Fulfillment (Amazon / B2B)
Each stage has its own inventory state and management requirements:
| Stage | Output | Key Concern |
|---|---|---|
| Farming | Raw harvest | Yield variability (weather, field size) |
| Cleaning | Clean product | Temperature & humidity control to prevent breakage |
| Packaging | Packaged product | Pack-ahead vs. pack-to-order timing |
| Fulfillment | Shipped inventory | Amazon visibility lag, shipment frequency |
The supply chain has a significant time lag between upstream decisions and downstream sales:
This means intelligence must flow upstream: market demand signals should drive planting volume decisions, not the other way around.
"The thing that you sell and the thing that you plant are connected. If you don't plant the right amount, you don't have the amount to sell. The intelligence flows upstream — your decisions about what you do are driven by the market."
— Mark Hope
The immediate goal is to answer: how much product should be shipped to Amazon, and how often?
Key questions to resolve:
- What is the optimal shipment frequency? (weekly, monthly, or event-driven)
- How much packaged inventory should be maintained at all times as a buffer?
- How do we account for the Amazon visibility gap — the window between shipping product and Amazon recognizing it in their system?
Planned tooling: A tool using the Amazon Seller API will pull current FBA inventory levels and recommend the next shipment quantity and timing based on sales velocity and stock-on-hand.
Rather than packing to order, Doudlah should maintain a standing packaged inventory target and pack continuously to replenish it. The model will define what that target level should be based on downstream demand forecasts.
Once the downstream model is stable, extend the intelligence upstream to inform:
- How much clean product to process and store
- How much raw product to send to the cleaner
- Ultimately, how much to plant in a given season
Because future sales are uncertain, the model will require scenario planning:
The model should surface the risk of each scenario — excess inventory vs. stockout — so Doudlah can make an informed planting decision.
Bean Vivo serves as a secondary sales channel and overflow outlet. Excess inventory can be moved through this channel at approximately $1.10/lb — a low-margin but non-zero recovery that prevents waste. This channel should be factored into the model as a floor on downside risk.