A manufacturer running both a B2B wholesale channel and a direct-to-consumer retail operation is, in effect, managing two different promises against the same physical stock. The B2B customer expects confirmed availability and reliable ship dates. The DTC buyer expects the item to be in stock when the website says it is. Both promises depend on the same underlying fact: how much inventory actually exists, where it sits, and whether it is genuinely available. When the systems recording that fact cannot communicate reliably with each other, both promises become unreliable at the same time.
TL;DR
- -Manufacturers fulfilling both B2B wholesale and DTC orders from shared stock face ATP inaccuracies and order failures when inventory systems update on different schedules rather than at the moment a fulfillment decision must be made.
- -Two approaches can reduce the gap: improving the integration layer so inventory events trigger cross-system updates rather than waiting for batch transfers, or adopting a four-tier data architecture with a governed tier that contextualizes and distributes inventory records to planning and order management applications.
- -Neither approach eliminates the complexity of managing inventory across heterogeneous systems and organizational boundaries; the semantic mediation layer requires ongoing maintenance as partner systems change, and the four-tier architecture requires upfront decisions about record ownership and data governance before it can be designed.
The Inventory Visibility Problem Across Channels
The core difficulty is not that manufacturers lack inventory data. A warehouse management system records physical movements at the distribution center. A retail point-of-sale system records sales and returns at the store. Each system generates its own version of inventory reality, and those versions may not be synchronized at the moment a fulfillment decision must be made.
A manufacturer's ERP may show available stock based on the last confirmed receipt or shipment, while the physical count at a retail location or return staging area has already moved in a different direction.
NIST research on supply chain visibility also notes that limited supply chain visibility remains a challenge, with companies still lacking sufficient insight into materials, components, and processes across their networks, even as investment in tracking technologies continues. The gap between what a system records and what physically exists is not a new problem, but it becomes more consequential when a manufacturer is simultaneously fulfilling wholesale orders with committed lead times and DTC orders where the customer expects immediate confirmation.
Why the Gap Persists
The architectural reasons behind this gap are well documented. The ISA-95 standard defines manufacturing systems across five levels, from physical processes through business planning, but the lack of integration across those levels has led to the manual or semi-automated duplication of data across all systems within the manufacturing enterprise. Each system ends up holding its own copy of inventory data, updated on its own schedule, with no guaranteed consistency across the set.
For inventory specifically, this means that the record a planner or order management system consults may reflect a state that existed hours or days ago. In a B2B context, that lag can produce an available-to-promise figure that overstates what can actually ship. In a DTC context, it can allow an order to be accepted against stock that has already been allocated, returned to a different location, or physically moved without a corresponding system update.
The problem compounds when inventory moves across organizational boundaries. Consider a manufacturer relying on a third-party logistics provider, a retail partner's distribution center, or a returns processor. The inventory event happens in a system the manufacturer does not own. The update reaches the manufacturer's planning or order management system only when the other party's system sends it.
Approaches to Closing the Gap
The architectural response to this problem involves reducing the delay and improving the fidelity of inventory data as it crosses system and organizational boundaries. Two broad approaches apply, and they are not mutually exclusive.
The first is improving the integration layer between existing systems so that inventory events trigger updates across the relevant applications rather than waiting for a scheduled batch transfer. The design-time work involves domain ontology definition, message annotations, message schema transformations, and reconciliation rules specifications, while the run-time layer performs the actual exchange and translation. This approach allows each system to retain its own data model while still participating in a shared inventory picture.
The second approach addresses the data architecture itself. NIST research on the digital thread in manufacturing describes a four-tier reference architecture designed to manage data generated by manufacturing systems, providing segregated access to internal and external clients while enabling the fusion of data across the product lifecycle and supply chain. In this model, a dedicated data management tier sits between the systems that generate inventory events and the applications that consume them, handling contextualization, transformation, and controlled distribution. Rather than each system connecting directly to every other system, the architecture routes inventory data through a governed layer that can apply business rules, resolve conflicts between records, and present a consistent view to planning, order management, and customer-facing applications.
Both approaches require deliberate design work. The semantic mediation research noted that the semantic annotation and reconciliation tasks proved to be challenging for the current toolset and for the users of those tools, and that significant rework of the tooling is needed for industrial use. A four-tier data architecture requires decisions about which system holds the authoritative record for each inventory state, how conflicts between systems are resolved, and who is accountable when the records diverge.
Tradeoffs Worth Understanding
Neither approach eliminates the underlying complexity of managing inventory across heterogeneous systems and organizational boundaries. The semantic mediation approach preserves existing system investments but introduces a translation and reconciliation layer that must be maintained as systems change. If a retail partner updates its inventory message format or a logistics provider switches platforms, the reconciliation rules need to be updated to match.
The four-tier data architecture can provide more durable separation between data producers and data consumers, but it requires upfront decisions about record ownership and data governance before the architecture can be designed. NIST research on supply chain traceability observes that individual traceability records should be structured to support specific decisions, and that this data must be presented in ways that multiple organizational functions can easily interpret and apply. That principle applies directly to inventory: the record needs an owner, a defined update cadence, and a clear path to the decision it is meant to support.
The practical tradeoff is between the change burden of redesigning integration layers and the ongoing cost of operating on stale inventory data. Batch-based updates between systems carry a lower initial change burden, but they perpetuate the gap between physical inventory states and the records that fulfillment decisions depend on. Tighter integration requires more upfront design work and ongoing governance, but it reduces the window during which a planning or order management system can act on information that no longer reflects physical reality.
Where to Start
Before selecting an architecture or integration approach, identify where the inventory record actually breaks down. Map the physical events that change inventory state: receipts, picks, shipments, returns, adjustments, and transfers. For each event, determine which system records it, how long the update takes to reach the systems that make fulfillment decisions, and whether any of those systems hold a conflicting record. That map will show where the gap between physical reality and system record is widest, and which boundary crossings carry the most fulfillment risk.
Once the gap is located, the question of whether to improve the integration layer, redesign the data architecture, or address record ownership first becomes answerable from the actual operating conditions rather than from a general preference for one approach over another. The architecture follows the operating design; the operating design follows the decision the record is meant to support.