Rainbow Department Store in China has quietly deployed Hanshow’s xPilot Digital Twin platform across its flagship locations, becoming one of the first retailers worldwide to operationalize AI-powered store execution at scale. The move, which leverages Microsoft Azure’s cloud and AI capabilities, was unveiled at NRF 2026 in New York, signaling a significant leap in how physical retail stores are managed, optimized, and personalized.

Hanshow, the Shenzhen-based provider of electronic shelf labels (ESLs) and retail IoT solutions, introduced xPilot at the National Retail Federation’s annual conference in January 2026. The platform integrates real-time shelf data, computer vision, and generative AI to create a living digital twin of the store floor. For Rainbow, a mid-tier department store chain with over 80 locations across China, the pilot directly addresses chronic operational pain points: out-of-stocks, planogram drift, and inefficient associate task routing.

From Digital Labels to Digital Twins

Hanshow built its reputation on ESLs—over 200 million units deployed globally—that wirelessly update pricing from a central system. xPilot extends that data pipeline into a full spatial model. Each shelf edge, product facing, and inventory movement gets mirrored in Azure Digital Twins. The system ingests feeds from Hanshow’s own cameras, third-party IoT sensors, and point-of-sale logs, then applies machine learning to generate alerts and recommendations.

“We’ve always known that the label is just the beginning,” said Hanshow CEO Hou Shiguang in an interview at NRF. “xPilot turns every shelf into a data source. It’s not about replacing staff—it’s about giving them superpowers.”

Inside xPilot: How It Works

The platform comprises three core layers. First, a perception layer uses Hanshow’s HiLPC (High-Precision Location and Computer Vision) cameras to capture images every few minutes. These feed object-detection algorithms trained on Rainbow’s product catalog, identifying missing items, misplaced facings, and pricing discrepancies with 98.4% accuracy in the pilot phase.

Second, a reasoning layer hosted on Azure Kubernetes Service processes the influx. It cross-references planogram databases, sales velocity, and even local weather forecasts to prioritize tasks. For example, if an ice cream brand is selling faster than restocking schedules allow, xPilot generates a “critical replenishment” task and pushes it to the nearest associate’s handheld device. The system also monitors out-of-stock durations; any gap persisting more than 20 minutes triggers an escalation to the store manager.

Finally, a generative AI interface lets staff query the twin in natural language. An employee can ask, “Show me all organic snacks running low in aisle 3,” and xPilot highlights problem zones on a floor map with recommended actions. The assistant is built on Azure OpenAI Service, customized with retail-specific prompts and a retrieval-augmented generation (RAG) pipeline drawing from Hanshow’s knowledge base.

Azure’s Role: AI, IoT, and Scale

Microsoft’s involvement goes beyond hosting. The xPilot architecture leans heavily on Azure IoT Hub to manage device telemetry from thousands of ESLs and cameras per store. Azure Event Hubs handle the real-time firehose of scanned data, while Azure Cosmos DB stores the constantly mutating graph of product-to-shelf relationships. For analytics, Hanshow adopted Microsoft Fabric, giving Rainbow access to Power BI dashboards that visualize store performance trends, planogram compliance scores, and labor efficiency metrics.

“This is exactly the kind of solution we envisioned when we expanded Azure Digital Twins for retail,” noted Microsoft Corporate Vice President of Azure Marketing, Alysa Taylor, during the NRF booth demo. “It connects physical operations to cloud intelligence without forcing retailers to rip out existing infrastructure.”

Hanshow also uses Azure AI Vision to detect shelf-edge conditions and Azure Machine Learning for demand forecasting models that adjust restock recommendations based on foot traffic patterns captured by in-store Wi-Fi analytics.

Rainbow’s Early Results

Rainbow began a three-store pilot in October 2025, expanding to 15 stores by late November. According to data shared at NRF, the early metrics are compelling:

  • Stockout incidents dropped 15.3% compared to control stores.
  • Planogram compliance improved from 82% to 96% within eight weeks.
  • Associate task completion time for merchandising checks fell by 22% because xPilot eliminated manual walk-throughs.
  • Customer satisfaction scores in pilot stores rose 7 percentage points, with shoppers specifically citing better product availability.

“Before xPilot, our staff spent hours each day physically scanning shelves,” said Rainbow’s CIO, Li Wei. “Now they act on exceptions. The AI prioritizes what matters, and the digital twin gives us a single source of truth across all our stores.”

Rainbow also tested xPilot’s generative AI-powered planogram authoring. Store managers can type a directive such as, “Create a healthy breakfast endcap for families,” and the system suggests a compliant layout based on sales history, margin data, and adjacent category affinities. Human merchandisers approve or tweak the suggestions; execution is then tracked via the twin to ensure fidelity.

The NRF 2026 Unveiling

At NRF 2026, held January 11–13 at the Javits Center, Hanshow positioned xPilot as the centerpiece of a broader “Store OS” vision. Attendees could walk through a mock produce aisle where digital twins updated in near real-time as products were moved. The demo highlighted how the system flags compliance issues—like a misplaced yogurt brand—and automatically reorders stock when sensors detect inventory below safety levels.

Hanshow also announced an ecosystem play: xPilot integrates with workforce management systems (Kronos, Reflexis via standard APIs), ERP platforms (SAP S/4HANA, Microsoft Dynamics 365), and robotics providers for autonomous restocking. A partnership with Simbe Robotics allows xPilot to dispatch Tally robots for shelf-scanning missions when camera coverage is sparse.

Expert Reactions

Industry analysts greeted xPilot with cautious optimism. “The concept of a store digital twin isn’t new, but Hanshow has the sensor density to make it work in a category-killer context,” said Gartner Senior Director Analyst Sandeep Unni. “Linking ESL data with computer vision and generative AI creates a compelling value proposition for grocers and department stores struggling with labor shortages.”

However, Unni warned about integration overhead. “Retailers need clean master data. If the planogram and inventory system records are inaccurate, the digital twin just mirrors garbage at speed. The pilot at Rainbow worked because they had already done the data hygiene.”

Forrester Principal Analyst Fiona Swerdlow added, “The key differentiator is the conversational AI layer. Most store execution apps are wizards or forms. Being able to ask ‘What needs my attention in dairy?’ and get a prioritized list is a usability leap. But enterprise adoption hinges on change management—managers accustomed to walking the floor may resist relying on a screen.”

Challenges and Considerations

Scaling such a platform is nontrivial. Each store generates roughly 2.5 terabytes of image and telemetry data per month, requiring robust edge computing and bandwidth management. Hanshow uses Azure Stack Edge appliances to process video locally, uploading only metadata and alerts to the cloud, which reduces latency and bandwidth costs.

Privacy is another dimension. While the cameras primarily face shelves, they can incidentally capture customers. Hanshow embeds facial blurring at the edge and retains images for a maximum of 72 hours unless flagged for training data (with opt-in consent management). Rainbow’s legal team conducted a privacy impact assessment aligned with China’s Personal Information Protection Law (PIPL), which xPilot passed without issue.

Cost is a barrier for smaller retailers. Hanshow declined to share exact pricing but indicated a per-store annual license in the “mid-five-figure range” for a typical 50,000-square-foot footprint, plus hardware. Rainbow’s deployment was partially subsidized through a co-innovation agreement with Microsoft, a model Hanshow hopes to replicate with other early adopters.

What’s Next for Retail AI

Hanshow plans to add predictive maintenance for store equipment (freezer temperature anomalies, HVAC performance) by spring 2026, folding those alerts into the same xPilot to-do queue. A sustainability module will track food waste reduction enabled by better inventory rotation, helping retailers report on ESG metrics.

Microsoft and Hanshow are also exploring integration with Azure Adaptive Cloud, allowing xPilot to run in hybrid deployments that span on-premises edge servers and multi-region Azure instances—an architecture that appeals to global retailers with data residency requirements.

The Rainbow success story is already attracting interest from other Asian and European retailers. According to Hanshow, negotiations are underway with a large Japanese supermarket chain and a German hypermarket operator, with go-live targets in Q3 2026. Both would represent the first non-Chinese deployments of xPilot.

As store operators recalibrate their technology bets in a post-pandemic landscape where labor efficiency and customer experience are paramount, the pairing of digital twins with generative AI offers a tangible path toward autonomous store management. With Rainbow now operating 15 xPilot-enabled stores and counting, the industry will be watching whether those early efficiency gains translate into sustained bottom-line impact.