The news: Target has been running a software copy of the system that decides how inventory moves from its warehouses to its stores. The company described the tool, called Proxima, publicly for the first time this month.
Proxima is a digital twin, a virtual model that uses the same data and rules as the live system. That lets teams test a change in the model before making it in the network.
The tool has two results to show so far. It predicted with 98% accuracy how product would flow through Target’s new Houston receive center before the building opened, and a fresh-food pilot across 63 items lifted on-shelf availability 2.5%.
COO Lisa Roath raised Proxima on the second-quarter earnings call as one of the tools behind Target's best item availability in years.
Target built the tool in-house after shopping the vendor market, and that decision is the part an operator weighing a twin can use. Three large US retailers have now disclosed network twins in 10 months, and they are splitting on build versus buy.
How it works: The middle mile is the part of Target’s network that moves products from its regional distribution centers into its stores. Jake Krings, Target’s vice president of technology for global supply chain and logistics, called it “the best way that we can influence” in-stocks.
Each positioning decision depends on lead time, the demand forecast, promotions, vendor schedules and transportation schedules. So changing one input shifts the others in ways that are hard to work out by hand.
Target had systems to see what was happening in the network and systems to act on it, but Krings said the company “saw the simulation capability as missing.” Proxima fills that gap using Monte Carlo analysis. The method runs the same plan many times with random variation in the inputs to show the range of outcomes a change is likely to produce.
The details: The first large test was the Houston receive center, a 1.2 million-square-foot building that opened in April. The building holds imported products until six regional distribution centers and a flow center need it.
Months before the opening, simulations showed the trailers leaving Houston would not be as full as the company wanted. In response, Target changed the inputs and logic until the model balanced trailer fill against store in-stocks. “Instead of learning as we went and reacting to things days and weeks after launch, we were able to anticipate exactly what was going to happen months in advance,” Krings said.
The second test was fresh food, where demand, lead time, perishability and transportation schedules all pull against each other. The food categories moved onto the new positioning system are showing a 2.5% lift in in-stocks.
Target built rather than bought because of category breadth. The company sells groceries, apparel, home goods, toys and electronics through one network. Krings said Target “didn't find a lot of packaged solution providers being able to support the breadth of our categories.”
The pattern: Target is the third large US retailer in 10 months to disclose a digital twin of its supply-chain network, and the three split on how they got one:
Walmart built its own. The company describes “virtual replicas of our logistics network” that it uses to simulate storms and other disruptions and pre-position inventory days ahead.
Lowe’s bought the stack. In October the retailer became one of the first companies on Palantir and Nvidia's combined platform, using it to create “a digital replica of its global supply chain network.”
Target built Proxima in-house on its own architecture.
Manufacturers have mostly gone the other way on plant-level twins. PepsiCo went to Siemens and Nvidia, where a pilot at select US plants and warehouses produced a 20% throughput gain. Unilever is rolling out more than 40 factory twins with Accenture over 18 months.
The two kinds of twins do different jobs. A plant twin models machines and line layouts, which transfer from one factory to another. A network twin models a company’s own inventory rules, vendor schedules and category mix, the part Target said vendors could not cover.
The counter: Target’s results come from Target. The 98% figure compares Proxima’s forecast with what actually flowed through Houston, and the 2.5% lift comes from a pilot of 63 items. So neither is large enough yet to show what the tool does across a network of nearly 2,000 stores.
In-stocks also improved for reasons beyond the twin. Target added $945 million of inventory in the first half, against $141 million a year earlier. And Roath credited pre-positioned seasonal inventory and dedicated trailer capacity for the quarter's availability gains.
In an Indago survey of 20 supply-chain executives in January, 40% said they were not considering digital twins, and only 10% were expanding beyond pilots. “I think digital twins make more sense the larger the organization in terms of physical locations and network flow points,” one respondent said.
My view: The Houston use is the one worth studying. Target found a trailer-fill problem months before the building opened and fixed it by changing inputs instead of changing trucks. A network-design test like that is where a simulation pays for itself before daily planning does.
On build versus buy, the deciding factor looks like category breadth more than size. A single-category retailer may find a packaged tool covers it. Target could not.
What’s next: Target plans to scale Proxima across more network functions and said the twin's output could eventually feed AI tools that evaluate options and automate some operational responses.
The next read on whether the in-stock gains hold comes with third-quarter results in November, when the fall store resets run into the holiday peak.
