Over the past year, our retail technology team at JD.com (also known as JINGDONG) has achieved remarkable progress in optimizing e-commerce operations through cutting-edge intelligent technologies. From supply chain fulfillment and advertising innovation to trade-in programs and customer service, we have elevated key aspects of our business to create seamless and exceptional customer experiences. Below, we share some of the standout highlights that demonstrate how these enhancements have driven value for both our customers and partners.

Comprehensive Upgrade of JD.com App

In the second half of 2024, the JD.com app underwent a major redesign and upgrade to provide an enhanced customer experience.

These improved features focused on creating better cognitive engagement, and a clearer homepage with a dedicated “My JD” section. On the homepage, customers can access JD.com’s comprehensive self-operated services, such as top-up, phone recycling, home cleaning, ticketing, laundry, auto services, express delivery and more, by scrolling down to the “second floor”. Meanwhile, key entries for deals, JD NOW (on-demand services) and new product channels are positioned on the first-level navigation for greater convenience. The new message center is designed to better meet users’ diverse needs, while product details have been optimized for more efficient decision-making.

Additionally, the application of new technologies, such as smarter search recommendations and large language models, ensures that users with varying needs can quickly find the products or services they require.

Intelligent Supply Chain Technology Powers Faster Fulfillment Timeliness

JD.com continues to be an industry leader in supply chain innovation. In 2024, JD.com became the only retail company from China named in the prestigious Gartner Supply Chain Top 25. This recognition highlights JD’s commitment to innovation and excellence in logistics.

JD.com operates a unique business model that delivers products directly to consumers through its self-owned logistics network, which includes over 1,600 warehouses and hundreds of thousands of in-house couriers. Its extensive product range is stored across a network of eight regional distribution centers (RDCs) in China, each managing millions of product varieties. To enhance delivery speed, each RDC is supported by several smaller forward distribution centers (FDCs) that are strategically located closer to customers. These FDCs primarily house high-demand products, with each capable of storing around 100,000 items. The selective storage capacity makes strategic inventory selection and allocation critical for fulfillment efficiency.

In 2024, JD.com’s retail technology team made significant strides in further reducing stockouts and fulfillment costs, to enable faster delivery for users. This was achieved by implementing a data-driven inventory selection and allocation algorithms, and by addressing issues around inventory balance across JD.com’s warehouses for over 10 million orders.

The system can analyze historical order data and co-purchase trends to create highly efficient product selection strategies. This helps maximize the fulfillment rate for “211” orders and enhances overall user satisfaction. The innovative end-to-end allocation algorithm integrates demand forecasting, iterative optimization, and simulation, enabling JD.com to deliver allocation plans that ensure the highest product availability at the lowest costs, all while providing clear and actionable insights for inventory management.

Thanks to a robust logistics infrastructure and intelligent technologies, more than 90% of JD.com’s retail orders are fulfilled within 24 hours, setting a new standard in the e-commerce industry.

In October 2024, the data-driven inventory selection and allocation algorithms technology was awarded the renown Daniel H. Wagner Prize at the INFORMS Annual Meeting for excellence in operations research and management science.

*211 or two-eleven indicates a standard shipping service for JD.com’s retail business: orders placed by 11 AM will be fulfilled on the same day, orders placed by 11 PM will be fulfilled the next day.

Large Models for Merchants to Generate Marketing Content

In the e-commerce landscape, the quality of product images and the precision of marketing content are critical factors that directly impact consumer purchasing decisions. However, creating these marketing materials can be a resource-intensive process for merchants.

To address this challenge, JD.com has developed its proprietary AIGC content generation platform, Jingdiandian (京点点). This innovative tool enables merchants to effortlessly generate high-quality product images, operational copy, short videos, and more, all at no cost with just a simple click.

Currently, these AI capabilities are leveraged over 10 million times each day, empowering more than 350,000 merchants to quickly produce a diverse array of marketing materials essential for effective store operations. This not only enhances content production efficiency but also significantly reduces overall production costs, allowing merchants to focus more on their core business activities.

Leading the Implementation of National Subsidy Distribution over Home Appliance Trade In

In alignment with national directives, JD.com has taken the lead in facilitating the distribution of subsidies for China’s home appliance trade-in program in 2024. Recognizing the regional and category-specific variations in subsidy policies, JD.com’s retail technology team swiftly established a qualification platform, creating a model to enhance efficiency across various businesses and cities. By utilizing a standardized national subsidy model, different business units across cities can now participate and complete the program in just a few minutes. Leveraging years of experience in AI and risk control strategies, JD.com has developed deep behavioral models to meticulously manage genuine users and identify high-risk users. This approach effectively prevents hoarding, arbitrage, and scalping, automatically intercepting over 99% of scalper users, with the proportion of risk users remaining below 0.5%.

Innovative Data Lake Technology Supports Real-Time Transformation of Trillion-Level Traffic Data

Within JD.com, data lake technology is rapidly evolving, providing robust support for the real-time processing and transformation of business data.

JD.com has selected Apache Hudi for its data lake, leveraging its unique business model to focus on I/O performance, feature richness, and ecosystem development. This has led to significant proprietary innovations, achieving multiple core features that surpass those of the open-source community. In 2024, JD.com successfully migrated its trillion-level traffic data assets into the data lake, reducing data latency from T+1 to minute-level latency, and achieving annual storage and computing cost savings of over 12 million yuan.

1-Second Response Time and 90% Decision Accuracy Intelligent Assistant for Merchants

JD.com’s Intelligent Merchant Assistant is designed to address the various challenges e-commerce merchants face in their operations by providing 24/7 operational support. Merchants can communicate directly with the intelligent assistant on the Jingmai (京麦) platform, whether they need to inquire about operational rules or execute quick function operations, with responses achieved in as little as one second.

The foundation of the merchant assistant’s algorithms is built on a multi-agent system based on large language models, simulating the collaborative operational methods of real-world e-commerce teams. This system effectively addresses the complex decision-making challenges merchants encounter in e-commerce operations, covering the entire process from product listing and order management to customer service communication and data analysis. It offers merchants functionalities such as sales forecasting, marketing deployment, pricing strategies, and opportunity keyword recommendations. Currently, the decision accuracy rate exceeds 90%, with individual system operations achieving response times as fast as one second.

Intelligent Advertising Solution Drives Reliable Creative Generation and Optimization  

In 2024, JD.com’s retail advertising team focused on creative generation and optimization, achieving technical breakthroughs in enhancing the usability of generated images, modeling real human feedback, and extracting creative representation information. These advancements effectively alleviate issues arising from data sparsity and the influx of massive creative inputs, enabling precise recommendations for online creatives and users.

In creative generation, the advertising team enhanced the usability of generated images by simulating human review processes through a multimodal robust feedback model. This significantly improved the images while maintaining visual appeal. The team also released the industry’s first RF1M (reliable feedback 1 million) dataset of human-labeled generated advertising images aimed at helping models more accurately reflect feedback.

For creative optimization, the advertising team utilized multimodal large language models to extract representative information from creatives to enhance the differentiation capabilities and cold-start performance. Furthermore, the creative optimization task has been divided into two phases: element selection and combination selection, allowing the optimization models to handle a richer array of creative materials.

Intelligent Shopping Assistant for Smart Shopping

JD.com has developed the Jingyan (京言) intelligent shopping assistant, an interactive application based on large language models, to provide users with comprehensive product information and professional recommendations through proactive service and multi-turn dialogue capabilities. The integrated technological architecture employed by Jingyan enables efficient search, product comparison, and review summarization across JD.com’s app, website, and mini-program.

To date, Jingyan has amassed over 50 million users. Customers can easily access the assistant by searching for “Jingyan” directly in the JD.com app or by clicking the Jingyan icon on the search results page.

Build An Immersive Shopping Experience by Leveraging JD.Vision and Naked-Eye 3D Solution

In June, JD.com launched “JD.Vision,” a native Apple Vision Pro app that offers an immersive shopping experience leveraging spatial computing, environmental sensing, eye-tracking, and gesture interaction technologies. It allows users to “drag and drop” their desired products into their home in a 1:1 scale, enabling them to visualize how each item will look in real-world space.

Building on this foundation, the team introduced the world’s first naked-eye 3D product marketing solution, known as the “Liying Plan.” Through the new function, shoppers can experience immersive 360-degree stereoscopic display of products by simply interacting with the JD.com app on their mobile phones, without wearing glasses or a headset.

(yuchuan.wang@jd.com)