A purchasing manager recently described a scenario that will sound familiar to anyone running MRP across multiple sites. Her system flagged a shortage on a resin used in one of their injection molding lines. The recommendation said to order 4,000 kilograms by Friday. She placed the purchase order, got it acknowledged, and moved on. Three weeks later, when the material arrived and the warehouse team tried to put it away, they discovered there was already 3,200 kilograms of the same resin sitting at the sister plant fifty miles away. Nobody had consumed it because nobody there needed it. The MRP at her site did not know the sister site existed in any useful way. The purchase order was technically correct. The business outcome was capital tied up in 7,200 kilograms of resin when 4,000 would have been sufficient.
This is the classic phantom reorder. The MRP is not lying. It is answering the question it was given, which is "do I have enough at this site to cover demand at this site within the horizon." The problem is that the question is wrong. In any operation with more than one location, the right question is "does the network have enough material, and if so, where is it, and can it be moved in time." MRP recommendation accuracy depends on whether your planning engine is asking the right question. If it is not, you will keep generating phantom shortages, placing duplicate purchase orders, and watching working capital accumulate at every site that happens to have local surplus.
The Siloed MRP Problem
Most MRP implementations were built on a foundation of single-site thinking. A plant ran its own MRP. Each plant had its own schedule, its own BOMs, its own stock records. If a second plant existed, it had its own MRP too. These systems were federated at best, meaning an analyst could log into each system and compare positions, but the planning engine itself could not see across the boundary. The assumption was that transfers between sites were exceptional and that each site was responsible for its own material position.
That assumption breaks the moment you have two sites running similar products with overlapping bills of materials. Suddenly the material you need at Site A is the material sitting at Site B, and the planning engine that only sees Site A has no way to account for it. Worse, if both sites are running independent MRP and both need 2,000 units of a shared component, both will generate a purchase recommendation for 2,000 units and you will end up receiving 4,000 from the supplier. The network's actual need might have been 3,000 with the remainder covered by one internal transfer.
Siloed MRP also creates a feedback loop that reinforces local surplus. When Site B has 3,200 kilograms of resin sitting idle, its own MRP sees that as healthy stock and does not act. Site A's MRP sees its own position as a shortage and places an order. The result is that Site B's surplus grows relative to its consumption rate while Site A's supply arrives from outside the network. Over months, this produces inventory distributions that are the opposite of what the total demand picture would suggest. You have too much at some sites, too little at others, and constant emergency reorders to paper over the imbalance.
Network Reservations Change the Equation
The fix is network-wide reservations. When a production order is confirmed, it should reserve materials against the network, not just against a single site's stock record. When MRP evaluates available-to-promise, it should consider the global pool of material minus global reservations, then distribute the remaining availability against the demand signals that drew it down. This is the shift from location-level MRP to network-level MRP, and it changes every recommendation the system produces.
FalOrb handles this through multi-level BOM reservations that span locations. When a confirmed production order at Site A requires 500 kilograms of a material, the reservation is visible across the entire network. Site B's surplus is not treated as unreserved stock; it is treated as potentially available for transfer to satisfy Site A's commitment. The MRP engine then asks the right question: given total network stock, total network reservations, and the ability to move material between sites, is there a net shortfall? If the answer is no, it recommends a transfer rather than a purchase. If the answer is yes, it recommends a purchase with quantity that reflects the true network gap, not the local gap.
This is also the mechanism that prevents duplicate purchase orders. With network-wide reservations, two sites cannot each generate an independent recommendation for the same material. The planning engine sees both demand signals, nets them against total network supply, and produces a single recommendation with aggregate quantity. A planner reviews one action instead of reconciling two. More on the architectural principle underlying this pattern is in the post about the immutable audit ledger, which explains why stock in a network-aware system must be derived from a single source of truth rather than edited in place at each site.
ATP Calculations Catch the Bottleneck
Available-to-promise is the finishing layer on top of network reservations. ATP tells you how many units of a finished product you can manufacture right now, given current material availability, existing reservations, and multi-level BOM requirements. In a single-site system, ATP is narrow and often optimistic. In a network-aware system, ATP is accurate and often uncovers the real bottleneck.
The real bottleneck matters because it changes what you buy. A production order for 1,000 units of a product might fail its ATP check not because you are short on the headline raw material but because you are short on a gasket that sits three levels deep in a sub-assembly. If the system only surfaces "ATP is zero," the planner has to dig into the BOM to find the cause. If the system surfaces "ATP is zero because we need 200 gaskets," the planner has an actionable signal and a specific recommendation. The post on available-to-promise as a factory floor metric goes deeper on this point.
When ATP calculations include network-wide availability, the bottleneck identification gets sharper. You no longer chase false shortages driven by local imbalances. You chase true shortages where the network as a whole does not have the material. This is where MRP recommendation accuracy stops being a theoretical problem and starts producing measurably lower procurement spend, lower working capital, and fewer emergency expedites.
Inventory Netting Across Horizons
Network reservations and ATP matter for current state, but the full picture of MRP recommendation accuracy also depends on how the system nets inventory across planning horizons. A recommendation that is correct for the 7 day horizon might be wrong for the 30 day horizon if the system is not projecting forward with reservation awareness.
FalOrb uses four configurable horizons: 7, 14, 30, and 60 days. For each item, gross requirement, scheduled receipts, and projected available balance are recalculated as new production orders are confirmed, new purchase orders are received, and stock movements occur. A shortfall in the 14 day window but not the 30 day window means the problem is timing, which is best solved by expediting an existing order or pulling a transfer forward. A shortfall in both the 14 and 30 day windows means the problem is quantity, which requires a new purchase order. The distinction between those two cases is completely invisible to a single-horizon MRP.
Netting also prevents phantom reorders that happen because the system double-counts demand. If a production order at Site A is already reserving material from Site B, Site B's projected balance should reflect the reservation. An engine that fails to net this reservation will see Site B as surplus and fail to act, while Site A sees itself as short and recommends a purchase. The material gets bought, the transfer never happens, and the surplus at Site B grows. Netting is the mathematical backbone that makes network-wide MRP work in practice.
How This Shows Up on the Screen
For a planner, the practical test of MRP recommendation accuracy is whether the recommendations on the screen correspond to actions that are genuinely worth taking. If the recommendation list is full of items you already have in the network, planners learn to ignore the list. If the list consistently surfaces real gaps and proposes the right action, planners treat it as a trusted input.
FalOrb's restock intelligence engine is built to generate trustworthy recommendations. It distinguishes internal transfer recommendations (one site short, another surplus) from reorder recommendations (net shortfall with no internal surplus) from redistribute recommendations (total stock sufficient but unevenly distributed). Each recommendation includes urgency, a plain-English headline explaining the situation, and a one-click action that either creates a transfer draft or a pre-filled purchase order. Recommendations auto-dismiss when the underlying condition resolves, so the list stays clean. Recommendations never auto-execute, so a human always confirms the action. Read more about the broader philosophy of this approach at falorb.com.
Closing the Phantom Reorder Loop
Duplicate purchase orders and phantom shortages are not MRP failures in the abstract. They are the specific outcome of asking the wrong question in a network environment. The right question is whether the network has enough material, and if so, where, and whether it can move in time. A planning engine that answers that question generates recommendations you can act on without second-guessing. A planning engine that does not answer that question generates recommendations you have to investigate before every purchase.
The cost of the investigation is usually hidden. It shows up as planner time, as working capital creep, as the occasional expedite fee, as the surplus pallet that nobody notices until the shelf fills up. It rarely shows up on a single report, which is why it persists. Making the shift to network-aware MRP, with true reservations, true ATP, and true netting across horizons, is how you stop paying that cost week after week. The operational discipline is the same. The system finally gives it something real to work with.
FalOrb helps manufacturers eliminate phantom reorders with network-wide reservations, ATP calculations that identify the true bottleneck material, and restock intelligence that distinguishes transfers from purchases. Book a 30-minute walkthrough or email us at [email protected] to see how it applies to your operation.