This Double 11, AI Is Busier Than Humans

This Double 11, AI Is Busier Than Humans

If last year’s Double 11 saw AI large language models (LLMs) in the e-commerce sector still in the exploratory stage, this year has entered a phase of large-scale implementation.

Tmall claims to have its first Double 11 with full AI implementation, launching 6 AI shopping guide applications covering the entire shopping process; JD emphasizes that this year’s Double 11 promotion is the one with the deepest integration of technology, with AI fully penetrating all core business links; Douyin also uses the Doubao LLM to create a new entrance in the e-commerce field.

Wang Yunfeng, CTO of Zhidemai Technology (35.550, 1.61, 4.74%), said in an interview with China Business Journal: “This year is not the first year of AI implementation in e-commerce. In fact, AI has been widely used in e-commerce before, but the capabilities of LLMs have undergone qualitative changes this year, allowing them to handle longer task chains, thus effectively expanding the breadth and depth of AI application scenarios.”

A Batch of AI Applications Launched

As the core hub linking consumers and merchants, e-commerce platforms’ AI applications mainly fall into two categories: one for consumers and the other for merchant operations.

On the user side, during this Double 11, Taobao launched 6 intelligent shopping guide tools such as “AI Universal Search”, “AI Help Me Choose”, “Photo Search”, and “AI Try-On”, covering the entire shopping process, aiming to improve search relevance and decision-making efficiency; Douyin uses the Doubao LLM to create a new e-commerce entrance—users only need to ask questions like “What to buy” or “What recommendations are there”, and Doubao will recommend multiple products, with links directly jumping to the Douyin Mall; the consumer decision-making platform “What Worth Buying” also enters with an AI-driven core strategy, whose AI shopping assistant “Xiao Zhi” has been upgraded to a new generation of consumer intelligence “Aunt Zhang”, with stronger perception, reasoning, and execution capabilities, providing users with intelligent services such as automatic price protection and wish lists.

It’s not hard to see that most e-commerce platforms have launched AI shopping guide applications for consumers. Wang Yunfeng told reporters that AI product recommendations have changed from the “trial stage” last year to the first choice for users’ shopping, “The attribute of AI as the entrance to users’ consumption decisions has become more prominent.”

On the merchant operations side, AI applications are more diverse. For example, JD’s intelligent customer service and digital human (18.870, 0.39, 2.11%) live streaming are both shopping tools for consumers and key drivers for merchants to improve efficiency.

According to JD, JD’s digital human JoyStreamer has served over 35,000 brands, with live streaming viewership exceeding 17 million so far, driving cumulative GMV over 700 million yuan; intelligent customer service has handled over 1.3 billion consultation services, covering all scenarios such as pre-sales, in-sales, after-sales, and logistics. In addition, JD has fully upgraded merchant-side tools during Double 11, covering store management, marketing promotion, service capabilities, etc., and added 20 free AI tools.

Tmall’s “AI Art Designer” generates an average of 200 million images and 5 million videos per month, increasing product click-through rates by 10%. “AI Data Analysis” has generated over 10 million reports, with 85% of merchants giving positive feedback. The AI customer service “Dian Xia Mi 5.0” helps merchants reduce costs by an average of 20 million yuan per day.

Kuaishou focuses on merchant operations, empowering merchants to improve operational efficiency with AI product capabilities from four dimensions: business opportunity insight, promotion strategy, production efficiency, and intelligent operation.

Weimeng’s intelligent operation Agent 2.0 for merchants also completed a new upgrade on the eve of Double 11, enabling all Weimeng merchants to use full AI capabilities. It not only strengthens AI personalized proposal capabilities on the planning side but also simplifies the full-link automation process on the execution side.

“What used to take 3 people 2 days to complete can now be done by 1 person in 1 hour, reducing manual costs by 80%,” said Sun Qian, head of Weimeng’s AI products. This year is a key turning point for AI applications from DEMO (prototype) to actual implementation, which is mainly reflected in the “breadth” and “depth” of applications. “Currently, AI has been widely implemented in multiple scenarios such as customer service, cross-channel product information updates, marketing planning, graphic creation, intelligent querying, and data interpretation.”

Significant Improvement in Model Capabilities

Wang Yunfeng believes that the large-scale implementation of AI in e-commerce in 2025 is due to the “significant enhancement of underlying LLM capabilities”, solving the “intelligence” problem of the models. “Faster inference speed, lower costs, and the addition of chain planning capabilities provide solid support for long-chain business scenarios.”

He gave an example to reporters: In 2024, many AI applications could only complete single-step actions, such as content review and simple AIGC generation (e.g., article summaries, product image production); by 2025, AI can handle 5-6 step long-chain tasks, covering more complex business processes such as digital human live streaming, automatic price verification, and high-quality marketing material generation.

Taking Zhidemai’s “good price verification” function as an example, in the past, manual verification was required to confirm the authenticity of “good prices”. Now, AI extracts structured information from exposure screenshots through multimodal models, automatically identifies unsubmitted coupons, then the agent simulates user behavior to verify the authenticity of the discount, and finally judges the applicable people, with no manual intervention throughout the process—only manual confirmation of the final conclusion is needed.

“This year, the multimodal capabilities of LLMs have significantly improved, enabling more accurate analysis of details in product images, such as text and materials, and one-stop generation of materials needed for e-commerce operations, such as videos and product posters. At the same time, AI is evolving from a single response capability to a unified intelligence with perception, decision-making, and execution capabilities,” Sun Qian pointed out. “Compared to traditional manual operations, AI has significantly improved processing speed and information consistency in tasks such as cross-channel product information updates.”

It should be noted that LLMs solve the core problem at the “intelligence” level, but for AI applications to truly land in e-commerce scenarios, targeted optimization is still needed.

“To improve the actual performance of AI in e-commerce scenarios, the team retrained the model by introducing operational expert knowledge, building a professional terminology library, integrating high-read social media content, and manual annotation. At the same time, it conducted compliance optimization in combination with the operational rules and industry norms of different platforms to ensure the professionalism and compliance of the output content,” Sun Qian added.

Opportunities and Challenges

Currently, vertical AI applications for different e-commerce scenarios are gradually emerging. In segmented operation scenarios such as data processing, copywriting, image generation, video production, shopping guide recommendations, and promotion planning, functional-specific AI tools have appeared, further adapting to the diverse business needs of the e-commerce field.

However, Cao Lei, director of the NetEase E-Commerce Research Center, pointed out that it should be noted that the differentiated strategies of various e-commerce platforms have invisibly increased the difficulty of tool selection for merchants. The AI tools launched by different platforms have their own focuses, and merchants need to accurately select them based on their business characteristics. Moreover, the controllability of technology application has become a major challenge facing the industry.

In Wang Yunfeng’s view, the core challenge of AI application landing in e-commerce scenarios stems from the extremely high requirements for information accuracy and timeliness in this scenario, coupled with the varied ways users express themselves, leading to the processing of massive unstructured data becoming the biggest bottleneck for current technology landing.

Massive data (15.780, 0.27, 1.74%) processing not only faces technical problems but also cost pressures. Sun Qian mentioned that in complex business scenarios of e-commerce operations, AI operation costs remain high, especially in high-precision image and video generation. In addition, there are problems such as poor cross-platform data collaboration and model adaptability differences in the e-commerce field. “For example, the application effect of the same AI function varies across industries, requiring targeted adaptation of technical routes for specific fields,” Sun Qian added.

From the merchant’s perspective, Cao Lei said that although head merchants have significantly benefited from AI tools, small and medium-sized merchants still face three practical resistances: high understanding costs, time required for popularization and landing, and insufficient cost-effectiveness. Currently, merchants’ applications of AI mostly stay at the basic level such as graphic generation and data insight. To achieve deep adaptation to complex industrial chains, it still takes time to polish.