When AI agents step into Yili's frontline services, a new solution emerges for in-store sales guidance and influencer marketing

A mother asked in the brand community, "My baby has a sensitive stomach. What should I pay attention to when choosing milk powder?"

To answer this question, the shopping guide needs to determine the baby's age and feeding situation, and then match them with the ingredients and target groups of different milk powders. If the consumer asks further questions, she also needs to know if there are any ongoing promotions, if the store has stock, and where it's most convenient to place an order. A slow response is an issue of efficiency, while an incorrect answer is a matter of trust. Wu Chunze, the general manager of Tencent Smart Retail's vertical industries, observed that "today's consumers are armed with AI and becoming more and more professional." Before asking the shopping guide, this mother might have read ten product reviews and compared the ingredient lists of three milk powder products. When she asks the shopping guide, she is not only seeking advice but also verifying the brand's professionalism.

This kind of pressure is being faced by front - line staff across the fast - moving consumer goods industry. Shang Zhihu, the general manager of Yili Group's Digital Technology Center, admitted that "in the past, it was difficult for a nutrition consultant to answer so many consumers' questions in a short time." For Yili, with a wide range of products, complex sales channels, and frequent promotions, tens of thousands of shopping guides face a large number of specific, scattered, and instant consultations every day, which further amplifies this pressure.

This is also the question that Yili began to ponder: Can the front - line service that used to rely on personal experience be transformed into a set of reusable capabilities with the help of AI?

AI Enters the Areas Closest to Consumers First

Yili first introduced AI to shopping guides, communities, and influencer marketing, areas where there is high - frequency direct communication with consumers.

The shopping guide scenario is the most typical one. In the community, a shopping guide has to face a large number of consumer questions every day: some ask about ingredients, some about promotions, and some about whether it's cost - effective to buy now. In the past, she had to find product information, look through promotion rules on her own, and then organize responses based on her experience. Yili has integrated product knowledge, promotion policies, and maternal and child knowledge into the AI agent. It's like having an always - online "expert" beside the front - line shopping guide.

In the past, the common content in the community might have been just "There's a promotion today. Take a look." Now, the shopping guide can quickly generate more targeted responses based on the solar terms, weather, consumer needs, and current promotion policies.

This change is even more significant in the maternal and child scenario. Shopping guides need to understand milk powder formulas, suitable ages, allergens, and feeding methods. It usually takes 3 - 6 months for newbies to get the hang of it. With the shopping guide agent, a newly recruited young shopping guide can also identify needs more quickly and give professional responses. Wu Chunze talked about the change: "It has improved the consumer experience. The consumer feels that when chatting with you, you understand him well and are very professional. The shopping guide has become more professional and caring because of AI, which has also brought in more sales and higher income for herself."

The changes are starting to show in the data: the click - through rate of product links in the shopping guide community has increased by 15.7%, and the number of shopping guide orders has increased by 26%.

The second obvious change occurred in influencer marketing. Brands need a large number of influencers to promote new products and generate sales. In the past, the biggest bottleneck was the efficiency of establishing connections. Business personnel had to find influencers one by one, send messages, discuss intentions, and arrange schedules. One person could handle at most two or three influencers a day. Yili developed an influencer marketing agent that takes care of everything from establishing connections, sending cooperation policies, determining intentions to standardized follow - up. Influencers with clear cooperation intentions are then handed over to real people for in - depth communication and implementation.

In these scenarios, AI performs repetitive, standard, and time - consuming tasks, while the parts that really require judgment and service are still handled by humans. In Shang Zhihu's view, "The fundamental change brought by AI is the extension of human capabilities and creative space. Suppose a person's productivity was 100 before. After using these agents, it might become 2000."

Pre - job Training for the Agent

When talking about how Yili selects AI implementation scenarios, Shang Zhihu emphasized "consumer - centric and business - oriented." If an individual uses AI to write copy and make summaries, personal efficiency will improve. However, enterprises are more concerned about whether this efficiency can translate into consumer value, employee value, and business results.

Based on this principle, Yili started its AI layout earlier than most enterprises. In 2021, Yili predicted that generative AI would become a strategic - level technology and targeted three directions: AIGC content generation, vibe coding, and intelligent customer service. At that time, ChatGPT hadn't become a national topic, and most enterprises didn't have as strong a perception of generative AI as they do today. Shang Zhihu said, "It's not a passive response to emergencies. Instead, it's more about our analysis and prediction of technology and a relatively proactive exploration."

With the direction set, the real challenge lies in implementation. Wu Chunze compared the enterprise agent to "a newly recruited digital employee." It naturally has search and generation capabilities and can master a vast amount of knowledge, but it lacks an understanding of Yili. With so many products, channels, and pieces of knowledge, how can it learn and apply them?

The first step is to organize the knowledge base. Product formulas, target groups, promotion policies, membership benefits, and compliance statements were previously scattered across various systems and stored in the minds of excellent shopping guides and operators. The same product might have different names in different systems, and the same promotion policy might have different rules in different regions and channels. Both parties spent a lot of time standardizing the information so that AI could accurately identify and call this knowledge.

The second step is to define the boundaries. Shang Zhihu summarized this as the relationship between "decision - making" and "strategy": "The agent mostly provides strategies and serves as an assistant. In important internal business processes and in line with institutional requirements, humans should make the final decisions." Wu Chunze also emphasized that the model cannot "run wild on its own." If it's compared to a horse, it needs good harnesses and guidance to run in the right direction in an orderly manner.

Only after the knowledge base is organized and the boundaries are defined can the AI agent be considered ready for work. The next step is to let it operate stably in real - world business.

Let the Agent Run Steadily

If Wu Chunze were to summarize the essence of the agent's work, his answer would be: "You need to connect to the right data and execute the right actions."

Tencent Cloud's Agent Development Platform (ADP) does exactly this. Wu Chunze described it as "a factory for agents." On this platform, enterprises can define agents, configure rules, and arrange processes, and then quickly move on to pilot verification. Yili can then focus more on business judgment.

For the agent to truly enter the business process, the knowledge base, process rules, permission boundaries, and system interfaces need to be connected one by one. What Tencent Cloud does here is not just building a platform but also includes security authentication, compliance safeguards, and data backup. Moreover, since most of the interactions between Yili's shopping guides and consumers take place in WeChat's ecosystem, such as enterprise WeChat and mini - programs, the agent can be more naturally integrated into the front - line workflow.

In the past few months, both parties have held many brainstorming sessions. The core of the discussions has always been business issues: What are the business goals? Which links need to improve efficiency? Which actions can be entrusted to AI? And what indicators should be used for verification in the end? Tencent Cloud brings its platform and industry experience, while Yili brings real - world scenarios and front - line feedback. The two parties work together to screen, verify, and adjust. Wu Chunze described this co - creation state as "make bold suggestions and verify them carefully."

Currently, about 40,000 employees at Yili are using AI. There are more than 2,000 L1 - level agents and dozens of production - level agents. Next, Yili aims to expand from several key scenarios to more comprehensive organizational intelligence.

Going back to the question from the mother at the beginning. The answer she gets today is supported by a whole set of organizational capabilities amplified by AI. The significance of this goes beyond efficiency itself.