IN Brief:
- Proxima models Target's middle-mile inventory positioning system using the same data and logic as its operational platform.
- A 63-item fresh-food pilot improved on-shelf availability by 2.5%.
- Modelling for the Houston Receive Center achieved approximately 98% accuracy before the facility opened.
Target is using a digital twin called Proxima to test inventory decisions before applying them to its live supply chain, giving planners a controlled environment in which to model how stock moves between facilities and towards stores. The system has already supported the opening of Target’s Houston Receive Center and a fresh-food pilot covering 63 products.
Proxima models Target’s middle-mile inventory positioning system, the part of the network responsible for determining how inventory moves between facilities before reaching stores. It uses the same data and logic as the retailer’s operational inventory platform, allowing proposed changes to be tested against a representation of the network without first altering live processes.
That distinction gives the system a practical role beyond conventional supply-chain visibility. Dashboards can show current inventory, demand, or capacity, while a sufficiently accurate digital twin can test how a proposed intervention changes those conditions. Inventory allocation, facility routing, replenishment rules, and network changes can therefore be examined before stores or distribution operations absorb the consequences.
Target used Proxima ahead of the opening of its Houston Receive Center to emulate inventory flows into and out of the facility. The company says the model achieved approximately 98% accuracy, allowing teams to compare planned flows with the behaviour expected once the building entered operation. That provided an opportunity to identify problems while the facility was still being prepared rather than after stock began moving through it.
A second deployment focused on fresh food, where replenishment is constrained by both lead times and expiry dates. Target tested changes across 63 items and reported a 2.5% improvement in on-shelf availability after using Proxima to simulate adjustments to inventory flow and correct issues before launch.
The size of the pilot is limited, but fresh inventory provides a demanding test because excess stock and insufficient stock carry different costs at the same time. Under-ordering can create shelf gaps, while over-ordering raises waste and working-capital exposure as products approach expiry. Changes that appear beneficial in a conventional demand forecast can therefore produce unintended effects elsewhere in the replenishment cycle.
Digital twins are increasingly useful where supply chains contain several interconnected constraints. A decision to move more stock through one facility can increase transport demand, alter labour requirements, consume additional storage capacity, or change replenishment timings downstream. Modelling those effects before implementation can reduce the amount of operational learning that has to take place after a network change has already gone live.
The Houston application also shows how the technology can support physical logistics investment rather than sitting separately as a software project. Bringing a new receive centre into a distribution network changes supplier destinations, inbound routing, facility capacity, downstream allocation, and transport flows. Even where the building itself is complete, the inventory rules around it still have to be tested against the wider system.
Model quality remains the central constraint. A digital twin that no longer reflects operating rules, facility conditions, or demand behaviour can create confidence without accuracy, making continued comparison with real results essential. Target says it intends to compare Proxima simulations with actual outcomes as the system expands across more network functions.
That validation will become more important if Proxima is eventually used to support AI-driven decision tools. Target has identified future agentic decision support as a possible development, with simulation outputs potentially helping systems evaluate options, prioritise actions, and automate selected operational responses. The current deployment does not amount to autonomous supply-chain management, and the distinction matters: Proxima presently provides an environment for testing decisions rather than replacing the teams making them.
The published results give the programme a firmer basis than a generic artificial-intelligence announcement. Approximately 98% modelling accuracy for a facility launch and a 2.5% availability improvement across a defined fresh-food pilot are measurable outcomes, although neither establishes how the system will perform across Target’s entire network.
Scaling the technology will expose more complex questions. A model used across additional facilities, product categories, and transport flows has to absorb far more variables while remaining fast enough to support practical decisions. The same-data architecture should help reduce the gap between simulation and live operation, but accuracy will have to be demonstrated repeatedly rather than assumed from the first two deployments.
The operational value will ultimately depend on whether Proxima reduces disruption and improves inventory decisions consistently as its scope grows. Supply chains rarely lack data; the harder problem is understanding the consequences of changing one part of a connected system. Target is using the digital twin to move that experimentation away from live inventory, where mistakes become physical stock problems, and into a model where they can be identified before deployment.
