Wumart Xueqing Road store, Beijing, China

Company overview

Wumart Group

Founded in Beijing in 1994 by Dr. Wenzhong Zhang, Wumart opened its first store, also recognised as Beijing’s first modern, standardised supermarket.

Over the past 30+ years, Wumart has grown into one of China’s largest full-channel, digitally enabled circulation and retail groups, and has been listed in the “China’s Top 500 Enterprises” for 24 consecutive years. Wumart operates 1,500+ stores across multiple formats nationwide and serves nearly 3 billion customer visits annually. Its businesses span retail, e-commerce, internet/IoT technology and logistics, and it continues to drive efficiency and customer experience through digital and technology innovation. As an industry pioneer, Wumart is now reshaping core operations with AI, accelerating the sector’s shift from “digital” to AI-led retail operations.

Dmall

Founded in 2015, Dmall (02586.HK) is a AI-driven retail technology provider, helping retailers transform from digital workflows to AI-led operations. Combining deep retail expertise with advanced AI capabilities, Dmall enables supermarkets, convenience stores, and multi-format retail chains to modernize core operations across merchandising, supply chain, store execution, customer engagement, and headquarters management.

Today, Dmall serves 438 retail clients across 10 countries and regions, empowering industry leaders such as Metro, Lawson, DFI Retail Group, and Pang Dong Lai.

Previous Next

Set the Scene

In early 2025, Wumart transformed its Beijing Xueqing Road store into a pioneering AI-Powered Flagship. Powered by Dmall’s intelligent infrastructure, the store seamlessly embedded AI into its daily execution loops. This unlocks smart assortment, dynamic pricing, automated replenishment, and real-time operational monitoring with proactive intervention, officially upgrading management from intuition-led to data-driven, intelligent collaboration.

Crucially, this programme is not a superficial tech overlay but a systemic reconstruction of the classic retail triad: People, Product, and Place. By threading AI throughout the entire value chain, the flagship perfectly synchronises operational efficiency, premium customer experience, and measurable ESG objectives.

Innovation Overview

Over the past decade, the retail industry achieved basic core workflow digitalisation. However, physical stores remained trapped in a reactive cycle, relying heavily on fragmented human experience and merely reacting to issues after the fact. To overcome this bottleneck, Dmall and Wumart co-created the “AI-Powered Retail” ecosystem, officially propelling the industry into the “AI Agent Era.”

Powered by Dmall’s extensive industry know-how and massive datasets from serving hundreds of retail giants, we engineered an autonomous, closed-loop optimisation ecosystem where AI acts as digital employees that can perceive, decide, and execute tasks:

• AI Merchandise Agent: Covering the end-to-end product lifecycle, it drives a perfect closed loop of key actions, from market insights and smart assortment to auto-replenishment, dynamic pricing, and slow-seller elimination.
• AI Store Agent: Acting as the store’s intelligent central nervous system, it provides 24/7 real-time monitoring of operational status, proactively analysing data and directly intervening in execution workflows.

This innovation fundamentally upgrades store management from “passive response” to “proactive prediction and dynamic intervention,” transforming fragile human intuition into calculable, replicable system capabilities:

• For Customers: Delivers a frictionless, highly consistent, and value-for-money shopping experience.
• For the Enterprise: Exponentially boosts merchandising power and labour productivity whilst slashing waste and energy consumption, perfectly underpinning rigorous ESG commitments with quantifiable results.

Challenge & Opportunity

The Challenges: Decision Gaps and Execution Bottlenecks

• Merchandising: Data-Rich but Insight-Poor.

In core scenarios like assortment, procurement, pricing, and slow-seller clearance, stores still rely heavily on staff intuition and guesswork. Retailers possess massive, dormant datasets but fundamentally lack the systemic capability to translate this information into optimal commercial decisions that automatically drive closed-loop frontline execution.

• Store Operations: Slow Responsiveness and Fragmented Execution.

Under the traditional model, essential daily tasks—including planogram management, floor inspections, replenishment, and customer service—are entirely reliant on manual physical intervention. This inherently reactive approach results in highly fragmented execution and severely delayed response times. Ultimately, it creates an insurmountable ceiling on staff productivity and compromises the consistency of the in-store experience.

Partnerships & Ecosystem Collaboration

This project transcends a conventional software rollout. It represents a deep, strategic co-creation between Wumart and Dmall. Through a rigorous mechanism of “joint design, joint validation, and continuous iteration,” the partnership successfully transforms pilot innovations into highly replicable operational capabilities.

• Wumart opened up real store scenarios and high-quality business data, sets commercial objectives and operating standards. Ongoing on-site validation provided traceable feedback loops that accelerated product and model improvement.
• Dmall supplied the underlying digital platform and AI capabilities, leading model training, strategy configuration and algorithm iteration. Crucially, by leveraging profound industry know-how accumulated from a global ecosystem of nearly 500 retail clients, Dmall adeptly analyses regional consumer nuances, ensuring AI strategies are hyper-localised for agile, actionable delivery.

Setup & Early Experiments

Since 2015, Dmall and Wumart have built a long-term digital partnership, starting with O2O integration and digitising the end-to-end retail operating model. As AI adoption accelerated across retail in 2023, they set a clear mandate: move physical stores from “digitalised” operations to a truly AI-native model.

Fresh food was chosen as the first pilot because it is both the most profit-sensitive and the fastest to validate. With short shelf life and high volatility, stores typically struggle with expiry management, delayed markdowns and high waste. Fresh is also data-rich and fast-turning, enabling rapid test-and-learn cycles. We therefore started in fresh with AI-enabled dynamic pricing and clearance, where optimisation impact could be measured quickly and credibly through two hard outcomes: waste reduction and gross margin performance.

Digital & Data Enablement

Data functions as the operational engine of the store, enabling AI to sense, decide, and optimise real-world retail operations. Massive, continuously accumulated high-quality industry data forms the foundation that allows AI models to generate reliable and actionable insights.

At the Wumart Xueqing Road flagship store, data powers the entire operational loop. By harmonising multi-source data—spanning transactions, inventory, equipment, and energy consumption, the store built a continuous cycle of: Collection ➔ Decision ➔ Execution ➔ Feedback ➔ Optimisation.  AI-driven strategies are dispatched directly into frontline workflows, while execution results flow back in real time to continuously refine models and optimise operational decisions.

Consider AI-driven replenishment: Analysing historical sales and product metadata against strict constraints like order cycles, inventory thresholds, and planogram capacity, the system precisely forecasts demand. It intelligently integrates variables such as seasonality, public holidays, promotional events, and weather patterns to dynamically optimise order frequencies and quantities. This proactive approach effectively eradicates out-of-stocks and capital-draining overstocking at the source.

Scaling the Innovation

To support agile expansion from a single-store pilot to a full-network rollout, we executed a systemic reconstruction across three dimensions:

• AI Agent Capabilities: Establishing the AI Agent as the intelligent hub, we transformed high-frequency decisions (assortment, replenishment, clearance, etc.) into a closed loop: Strategy Generation ➔ Task Dispatch ➔ Execution Feedback ➔ This reduced reliance on individual judgement and improved consistency across stores.
• Organisational Evolution: Transitioning staff from passive “system operators” to active AI co-workers. By redesigning role divisions and implementing a “rapid-cycle training + on-site coaching” model, staff understand the “why”, “how” and expected standard behind AI-guided actions, enabling reliable human–AI collaboration in daily operations.
• Adaptive Localisation: On a common platform and standard modules, stores were parameterised by format, customer profile, size and in-store equipment. We also adapted layout and traffic flow to protect experience quality as capabilities scaled.

Operational Transformation

This innovation shifts store operations from experience-led, reactive management to AI-driven, system-orchestrated daily execution.

1. Merchandising: The AI Merchandise Agent digitises and productises the unstandardised, tacit experience of top-tier buyers (e.g., origin tracing, seasonal windows). Transforming this fragile knowledge into scalable industry know-how, it elevates high-frequency actions—auch as assortment, replenishment, pricing, and clearance—into AI-calculated strategies that directly drive frontline execution. Product sell-through remained consistently above 90%, while inventory turnover improved to around 20 days, significantly accelerating stock rotation.
2. Operations: By eradicating reliance on inefficient manual floor patrols, the AI Store Agent provides real-time, terminal-led sensing of shelf and operational statuses. Upon detecting anomalies—such as out-of-stocks, planogram non-compliance, or near-expiry risks—it instantly dispatches standardised task directives. This synchronises unprecedented labour productivity with elevated in-store service, guaranteeing flawless execution consistency. The system significantly reduces manual inspection workload and improves labour productivity, contributing to an overall 30% reduction in labour costs.

Setbacks & Pivots

Early on, the biggest obstacle was not technology. It was mindset and working habits. Frontline teams were used to experience-led decisions and initially viewed AI with caution or resistance, including concerns about being “replaced”. At the same time, cross-role and cross-department coordination was not yet strong enough, which led to variation in how strategies were executed on the floor.

To overcome this, we clarified roles and SOPs and converted AI outputs into concrete, executable tasks, along with rapid-cycle training, on-the-floor coaching and weekly reviews. Once teams saw that AI reduced unnecessary effort, improved shift efficiency and delivered better commercial outcomes, attitudes moved from scepticism to active adoption, with staff proactively suggesting improvements.

The key learning was clear: digital transformation is not installing software, but redesigning routines and collaboration. Only when the decision–execution–feedback loop is embedded into daily work can AI scale sustainably.

Impact & Results

1. Financial growth

• Average daily in-store sales increased by 258.7% post-transformation.
• In-store transaction volume rose by 144.7%.
• Average basket value increased by 46.6%.

2. Operational & supply chain efficiency

• Product sell-through remained consistently above 90%.
• Inventory turnover improved to around 20 days, indicating materially faster stock rotation and healthier inventory flow.

3. Customer experience & membership

• Customer satisfaction reached 95%+.
• Approximately 8 million new members were added in three months.
• Active members contributed about 85% of sales, demonstrating strong loyalty and engagement.

4. ESG, risk control & cost optimisation

•  Intelligent energy management reduced overall energy consumption by 25.9%.
• Labour costs decreased by approximately 30%.
• AI-enabled loss prevention reduced self-checkout missed scans and shrinkage by around 85%, strengthening margin protection while supporting a smoother checkout experience.

Scalability & Future Potential

Proven rollout within Wumart

Using the Xueqing Road flagship store as the reference blueprint, we packaged the validated “AI-powered retail” capabilities for rapid configuration across different store formats (hypermarket, community store, etc.). The model has already been replicated across 77 Wumart stores, expanding from a single-city proof point to multiple provinces and market tiers in China, demonstrating repeatability and operational stability across regions and formats.

Dmall’s roadmap: full “agentisation” and broader co-innovation

As the technology partner, Dmall is advancing towards a comprehensive AI ecosystem and scaling this AI-native operating model to more retailers globally through co-creation and localised deployments.