Retail AI gets the most attention when it talks to shoppers. Target’s more consequential experiment is happening somewhere customers will never see it: in the middle of the supply chain.
The retailer has built Proxima, a digital-twin system designed to model how inventory moves through its network before Target changes the real thing. It is less cinematic than a humanoid robot or a generative shopping assistant, but it targets one of retail’s most expensive problems — making thousands of interdependent inventory decisions without learning every lesson in production.
In a small pilot involving 63 fresh-food items, Target says Proxima let teams simulate changes to inventory flow, spot problems before launch and improve on-shelf availability by 2.5%. For its Houston Receive Center, Target used the twin before opening to model inventory flows with about 98% accuracy.
Those percentages deserve context. A 63-item test is not proof that a digital twin can optimize an entire national network. But shelf availability is a brutally practical metric. Small improvements, repeated across a large assortment and thousands of stores, can matter far more than an AI feature that generates a clever product description.
The system also points toward a more credible version of “agentic retail.” Target says the insights generated by Proxima could eventually support AI-powered decision tools that evaluate options, prioritize actions and automate operational responses.
That sequence is important. An agent is only useful if it has a trustworthy model of the system it is being asked to change. In supply chains, a confident but wrong autonomous decision can create stockouts, excess inventory or cascading capacity problems. A digital twin gives AI somewhere to rehearse.
Target is effectively building a sandbox for operations.
That does not eliminate the hard work. The twin has to remain calibrated against reality, and fresh food, apparel, seasonal merchandise and general merchandise all behave differently. Models will drift when demand patterns, suppliers or network topology change. Human operators still need to know when the simulation is wrong.
But the direction is notable. Retailers spent the last decade collecting more real-time signals from stores, warehouses and transport networks. The next competitive advantage may come from being able to test decisions against those signals before committing physical inventory.
If that works, the most valuable retail AI may be the AI that prevents a bad decision before a customer ever knows one was possible.