AI technology is accelerating its penetration into the e-commerce field, from AI shopping guides, digital human live broadcasts to intelligent customer service, major e-commerce platforms have been deployed, trying to improve user experience and operational efficiency through technology. However, despite the many new scenarios and possibilities that AI has brought to e-commerce, consumers do not seem to buy it. From the mechanical reply of AI customer service to the inaccurate recommendation of AI shopping guides, the application of AI in the field of e-commerce always seems to be unsatisfactory.
The entry of AI into e-commerce is nothing new.
For platforms, AI is becoming a new quality productivity for e-commerce. New scenarios such as AI shopping guides, digital human live broadcasts, and AI dress-up changes have been put into their own apps by major e-commerce companies. While reducing costs and increasing efficiency, it may be possible to gain new user experience possibilities.
However, the good possibilities are still too far away. For C-end consumers, the experience is not necessarily as silky as the platform imagines.
The deepest impression of ordinary people on the application of AI in e-commerce is probably the automatic reply of the robot customer service, which not only can’t understand you, help you solve the problem, but even ask you to follow the dialogue format it wants.
As a result, you can only get emotional, your tone deteriorates, and strongly ask for “labor transfer” one sentence after another.
Whether the layout of AI by platforms such as JD.com, Taobao, and Douyin really meets expectations, may be able to find some answers from consumers’ physical experiences.
Do C-end users find the AI e-commerce feature easy to use?
For ordinary consumers, AI’s entry into e-commerce seems to be an area worth exploring, just like an open-world level, if the rewards in it are not attractive enough, players will not take the initiative to explore.
For example, mainstream scenario AI shopping guides are being built by mainstream e-commerce platforms.
What does a product manager need to do?
In the process of a product from scratch, it is not easy to do a good job in the role of product manager, in addition to the well-known writing requirements, writing requirements, writing requirements, there are many things to do. The product manager is not what you think, but will only ask you for trouble, make a request:
View details >
In May, a good friend of Xiaoyu celebrated his birthday, and after discussion, several friends decided to buy a gift worth less than 3,000 yuan. Everyone discussed several options, but couldn’t make up their minds. Xiao Yu recommended “Taobao Ask” by a friend, wanting to see if AI can give new suggestions. This is an AI shopping guide function that Taobao tested internally in September 2023.
After searching for a long time, she didn’t find “Taobao Ask” on the Taobao homepage, so she had to enter it manually, and then entered the chat dialog box with the virtual character “Xiao Tao”.
Similar to the chat box she usually uses Deepseek and ChatGPT to talk to, the difference is that “Taobao Ask” will “preemptive”: it came up and posted a text, guessing that Xiao Yu has been particularly interested in aromatherapy and household products recently, and gave some suggestions.
However, before a friend got married, Xiao Yu sent an aromatherapy, probably because of the previous browsing records, so Taobao gave such a suggestion. Xiao Yu said, “In the scene of gift-giving, I may need to frame a range first, and then choose, instead of dying in a category at the beginning.”
But in the view of another Taobao user, Bubu, these suggestions are not fundamentally different from the “guess you like” on the homepage.
So Xiao Yu returned to the ignored “guess what you like” and directly asked “gifts for young girls within 3,000”, but was amused by “Xiao Tao’s answer: “Hey, giving gifts is a technical job! Do you want to make her happy or moved?” Xiao Yu, who is used to using ChatGPT, complained that “Xiao Tao” is like a nascent version of AI, with a kind of silliness that does not understand the world.
Usually, Xiao Yu usually uses Xiaohongshu to search for gift-giving strategies, or goes to Douban’s “A gift for you/Gift Center” and other groups that specialize in discussing gifts, and many bloggers will sort out gifts in a specific price range, or give suggestions for specific groups of people.
In contrast, Xiao Yu can feel that Taobao AI is working hard to do scene recognition, but at least in this link of gift-giving, she can feel that the AI’s recommendation lacks rules, or it is a single category, such as only recommending jewelry to her, but the price is too different; Or it has a bit of a stereotype, such as skirts and cosmetics for girls, “in fact, most of the things we choose ourselves are electronic products.”
What AI cannot understand is the balance between budget and surprises for real migrant workers.
For Xiao Yu, “within 3,000 yuan” is about 3,000 yuan or less, assuming that the instruction is sent one less sentence, the AI’s understanding will be biased – Deepseek even recommended a gift of about 300 yuan to her, which is too low compared to her expectations. “In this comparison, Taobao also gave at least a dozen links to reasonably priced and high-quality products.”
“Maybe giving gifts is also a more laborious thing, and it is still difficult to expect Taobao AI to screen out satisfactory things through a few rounds of simple conversations.” Xiao Yu and his friends finally chose one of the gifts that had already been decided.
In the AI scenario, many consumers have encountered funny scenes of chickens and ducks, as well as a Q&A loop like “being haunted by a ghost”.
Someone asked Douyin’s “smart shopping”: What kind of gift is suitable for grandma? The other party seemed to be asleep and asked what the gender of the elder was.
Another feature with mixed reviews is the AI fitting.
It has actually appeared in the product links of Taobao and other e-commerce platforms. Many people are curious to experience Taobao’s fitting function, but they feel a little tasteless.
Xiao Jiang likes to buy clothes and cosmetics online, but after tinkering, time runs fast, and after tossing once, she doesn’t want to try again. “There is still a certain gap with the real person’s upper body, I feel that some clothes are a little inconsistent, but the upper body effect is okay, and the return and exchange is not particularly troublesome, it is all the couriers who come to pick up the goods.”
Perhaps because of the limited positive feedback from users, Taobao fitting is still in the partial testing stage.
In addition to fashionable technologies such as AI shopping guide and AI fitting, AI scenarios such as digital human live broadcast and one-click dress-up change are also gradually penetrating into e-commerce shopping pages, and of course, there are AI customer service that has become a joke on the Internet.
In short, the ambitions of e-commerce giants for AI are difficult to hide.
E-commerce giants are quietly making efforts
Most consumers may explore AI scenarios sporadically, with some vague somatosensory. From the perspective of e-commerce platforms such as Taobao, JD.com, Pinduoduo, and Douyin, the layout of AI is an indisputable thing.
AI technology has helped merchants reduce operating costs, improve work efficiency, and bring more considerable profit growth.
In Taobao, AI Agent-Quick Butler is like a small team, which is a single-digit virtual employee, but it is these few employees who can master the basic skills of e-commerce operation, intelligent data analysis, and optimize store operation strategies.
In addition, during last year’s “JD 618”, Yanxi digital humans helped about 5,000 brands start broadcasting stably, with a cumulative number of views of more than 100 million, and 21 president digital humans “debuted” in JD.com’s live broadcast room.
Kuaishou e-commerce followed suit and launched the ASR model this year, which can automatically identify the anchor’s oral content during the live broadcast and generate real-time comments for users who watch the live broadcast, and users only need to move their fingers to interact with the anchor.
The platform has tasted the sweetness, so it spares no effort, and from the perspective of the brand, embracing AI is also the general trend.
The “Make a Friend” live broadcast room was connected to DeepSeek at the beginning of this year, fully entering the intelligent stage. Huang Nan of the Ministry of Information Technology of Make a Friend mentioned in the official account that AI has greatly improved the efficiency of live broadcasting. Oral broadcast script writing, short video script disassembly and generation, subtitle optimization, compliance risk control review, many links have been intelligent.
This year’s 618, AI has become the main force of combat. “Taking Taobao as an example, 90% of the oral broadcast content in our live broadcast room is automatically generated by AI and manually calibrated, which greatly improves work efficiency.” Huang Nan said.
AI+ e-commerce, the ideal link is that the platform empowers merchants, merchants use AI technology to reduce costs and increase efficiency, and the saved production capacity can be used for new product research and development and other fields. In terms of actual effects, AI technology has indeed brought benefits to platforms and merchants. The “volunteer boss lady” that became popular at the beginning of this year is the best proof of accelerating the times and putting efficiency first.
In addition to tangible interaction, for consumers on e-commerce platforms, more often than not, AI is hidden behind it.
The e-commerce industry competes for traffic, and the emergence of AI models has improved efficiency and made it more eye-catching. After getting the traffic, it is necessary to speed up, improve efficiency, and iterate endlessly.
According to iiMedia Consulting, in 2025, e-commerce will account for 16.49% of the industry distribution of AI digital human enterprises in China, which is the largest proportion.
But does somatosensory keep up with efficiency?
In order to allow users to take a breather in the complicated price calculation, Taobao launched the “Tmall Bargaining Assistant”, and Pinduoduo also launched the “Automatic Price Following” system.
For consumers, using AI to bargain in a circular way basically only requires the movement of a finger. But for merchants, during the promotion, they have to deal with a large number of inquiries, and this system makes them “sweat”.
It is obviously an AI tool that improves communication efficiency and saves labor costs, but it has become the object of complaints, which exposes the weakness of AI e-commerce in the present – understanding data, but it seems that it does not fully understand people.
From data to people
According to iiMedia Consulting’s data, at present, consumers have a great demand for manual intervention functions, with 45.5% of consumers using them occasionally and 33.7% of consumers using them when needed, indicating that AI e-commerce still needs to improve its problem-solving capabilities in complex scenarios.
The platform is making efforts, merchants are also working hard, and consumers are trying, but why does AI+ e-commerce often feel a little tasteless?
Taking AI shopping guide as an example, merchants can significantly reduce labor costs by accessing large models. Some companies and merchants have taken big steps and laid off a large number of manual customer service, but in many scenarios, AI is actually not enough to cope with complex and changeable needs.
For example, in a widely used AI e-commerce customer service scenario, users are repeatedly asked to “describe the problem”, the AI cannot recognize the keywords, and the AI is still unable to do so after changing the statement, and there is a lack of supplementary solutions. When users send “manual transfers” many times, the AI at this moment shows a strong desire to survive and is unwilling, and is still fighting for the opportunity to make contributions, of course, this is because the technology sets parameters in the background. The posture of the AI’s efforts did not win understanding, but was tiresome. After thousands of calls for manual services, users may have to repeat another round of problems……
These are the “hard wounds” that AI cannot completely replace human customer service, and even if you access a full-blooded Deepseek, you may need to be well “tuned”.
Nowadays, a considerable part of the work of e-commerce customer service is to deal with after-sales problems, usually to this step, in addition to ordinary returns and exchanges, registered addresses and other basic services, that is, to send the wrong goods, receive defective products, etc., consumers have some emotions. In the report of “Shanghai Broadcasting Reporter Chat Box”, a consumer got the wrong product, complained, applied for after-sales, and was informed of the result of the processing was a cold AI phone call.
At the same time, everyone’s demand for emotional value is only increasing, and it is far from enough to rely on the basic words “pro” or “~” when Taobao first emerged. This also puts forward higher requirements for the AI large model accessed by merchants: to be able to understand the needs of users, predict the products that may be needed, and also to give emotional comfort, and to be careful.
But the current situation is that many e-commerce platforms are still completing the transformation from data to AI, and the next step is from AI to people, and in the future, AI+ e-commerce scenarios need to find a balance between more personalized in-depth needs and rapidly developing technologies.
In this regard, major e-commerce companies are also promoting technology iteration. For large factories, human resources systems that are good at solving problems in depth and rapidly iterative technology are huge advantages, so AI pain point solutions are often top-down, from platforms to merchants, from large factories to small factories.
Alibaba, which is AI-based in all its business, will already be implemented in a large number of e-commerce scenarios on Double 11 in 2024, empowering merchants in search, recommendation, and advertising, and improving the efficiency of delivery, product delivery, etc.
In the future, e-commerce AI will enter full-cycle operation from single-point functional iteration. For example, now that a user enters Taobao, Taobao’s big data can use algorithmic recommendations to give the option of “guess what you like”, and users can also have direct conversations with merchants, but for most users, this interaction is instantaneous and short-lived.
To continue this interaction, let users have a sense of trust, instead of resisting and irritable when they see AI robots, and need more in-depth and lifecycle operations, that is, when consumers enter, big data already knows what categories they have bought, what categories they may be interested in the future, and more importantly, AI should be more accurate in the next life cycle of users, and should make more recommendations on what aspects in the next life cycle, rather than solving the click-through rate problem of single-point clients.
Prosus&dealroom’s report predicts that in the future, AI will be an all-weather shopping butler that can cope with more and more complex scenarios. But at the same time, people’s needs will also change with more diverse scenarios.
For the fiercely competitive e-commerce track, the difference may be a complete failure. Perhaps, a better solution is not to pursue a one-and-done solution, and everything will be fine if you access AI, but to maintain a keen sense of human nature, so that AI can continue to iterate and continue to become a more powerful “gain magic weapon”.
