Apart from not being able to be a son-in-law, how is DeepSeek worse than Dong Yuhui?

E-commerce platforms are relatively cautious about accessing DeepSeek, and the reasons behind this are worth pondering. This article will explore the role of DeepSeek in consumer decision-making, analyze its differences from traditional e-commerce recommendation models, and the strategic choices of e-commerce platforms in the era of AI recommendations.

"DeepSeek recommendation" is becoming a new consumption decision-making ritual for young people.

Before opening the shopping app, Liu Xue, a white-collar worker working in Beijing, first sought help from DeepSeek: "I am 27 years old and have sensitive skin. Please recommend some sunscreens that are suitable for me."

A few seconds later, DeepSeek listed a detailed report and recommended multiple brands of sunscreen based on skin condition, ingredients, matching degree and other dimensions. In the end, Liu Xue bought one of the sunscreens based on her situation.

Chen Peng, 26, who works in product operations, wanted to broaden his horizons through reading, so he asked DeepSeek, "What are the ten most important books to read?" DeepSeek also gave him a list of books and classified them with reasons. Chen Peng chose to buy a few of them.

This used to be the job of Li Jiaqi and Dong Yuhui in the live broadcast room: shopping guide.

In the past few years, top anchors have built trust with users and established a product distribution mechanism centered on the anchors, recommending what they think is the best and most suitable things to their fans.

But now this recommendation system is being deconstructed by DeepSeek. The long thinking chain advantage of DeepSeek's deep thinking mode can provide users with more comprehensive, accurate and high-quality answers, and then form one-to-one personalized recommendations. When applied to the shopping field, it has become the D choice - DeepSeek's preferred choice.

The essence of “D Selection” is “AI Shopping Guide”, which assists users in making consumption decisions efficiently. The scenario of “AI Shopping Guide” is not unfamiliar.

Many years ago, some e-commerce platforms tried to use big models to achieve a closed transaction loop of "planting grass + shopping".

For example, Taobao's AI assistant "Taobao Wenwen" has long been connected to Tongyi Qianwen, and its functions include personalized recommendations and generating purchase suggestions. JD's Yanxi model is also connected to the consumer shopping guide scene, and "JD Jingyan" is also clearly positioned as an "exclusive AI shopping assistant." Douyin APP's AI search service is also connected to the Doubao model.

E-commerce platforms all want to do the "AI + e-commerce" business. However, e-commerce platforms have not given this function a prominent traffic position, and AI search has not appeared in the first-level entrance of mainstream e-commerce platform apps. For example, Taobao's "Taobao Q&A" and JD's "JD Jingyan" both require you to actively enter Taobao Q&A or JD Jingyan in the APP search box to enter the corresponding page.

On the other hand, due to various reasons such as technology and user habits, this function has not become the mainstream choice for consumers to shop.

The emergence of DeepSeek, which has stronger technical capabilities and is more widely discussed, is changing this situation.

The Daily Economic News quoted the data analysis platform QuestMobile as reporting that as of February 9, the cumulative downloads of DeepSeek's app have exceeded 110 million times, with the highest weekly active user scale of nearly 97 million. Among the huge number of users, some people have already used DeepSeek as their shopping assistant. "D selection" is becoming a new consumption ritual for young people and entering the online shopping process.

Some brands, such as Amul and Anker, also proactively put "DeepSeek Recommendation" on the main picture and use it as a marketing point.

However, facing the DeepSeek wave, e-commerce platforms are not very active. So far, except for Alibaba's 1688 merchant version, Taobao, JD.com, Pinduoduo, Douyin and other major e-commerce platforms have not announced the access to DeepSeek.

E-commerce platforms have an advantage in AI recommendations because they can directly attach shopping links to the recommended references to form a closed transaction loop. Users ask questions to the AI ​​connected to the e-commerce platform, and the AI ​​gives suggestions and attaches links with more favorable prices, thus eliminating the need for the anchor as an intermediate link. From search recommendations, personalized recommendations, content recommendations, anchor recommendations to the current AI recommendations, the recommendation methods of e-commerce platforms are undergoing a new migration.

01

"D selection" is becoming the first choice for young people before shopping. Liu Xue has searched for cosmetics suitable for her skin type on DeepSeek more than once.

"Hello, I'm 27 years old. Please recommend a skin care product for me. It has the following effects: oily skin, red cheeks, yellow T-zone, and uneven skin tone. The price is within 200-400 yuan."

In less than a minute, DeepSeek output a detailed report. From cleansers, essence water, to essences and lotions, DeepSeek recommended a brand for each product and listed the price, core efficacy, adaptation needs and other characteristics one by one.

More importantly, DeepSeek also sorted out usage suggestions such as how to zone and care for the skin. After reading the detailed report, Liu Xue decided to buy a brand of lotion recommended by DeepSeek.

Liu Xue once bought a lotion for sensitive skin from a live streamer, but after using it, she found that the product made her cheeks sting. In Liu Xue's opinion, "sensitivity" is just one of her skin problems. "The streamer couldn't take my entire skin condition into consideration when recommending products, but DeepSeek can do it one-on-one and recommend products based on my special skin condition, which is clearer than the streamer's explanation."

DeepSeek's analytical capabilities in the cosmetics vertical field have been recognized by professionals. Chen Huo, who works in product development at a cosmetics company, said that DeepSeek can help them analyze ingredients for the development of new skin care products.

On social platforms such as Xiaohongshu, many netizens have shared their experiences of using "DeepSeek recommended cosmetics", and these cases all show that DeepSeek can make precise recommendations based on user needs. For example, one netizen told DeepSeek about his/her usage scenarios and skin type, and asked DeepSeek to recommend lipsticks; another asked DeepSeek to recommend cosmetics brands based on his/her income and skin type.

DeepSeek's personalized recommendations are not only used by users in the areas of cosmetics, but also in the areas of snacks, clothing, 3C digital products, daily necessities, etc., and DeepSeek can be used for help.

For them, one of the advantages of DeepSeek recommendations is one-to-one accurate recommendations; the second is objectivity and truthfulness.

A Xiaohongshu netizen who used DeepSeek to select a refrigerator shared, "Using DeepSeek to assist with shopping is absolutely amazing. I am not afraid of offending the manufacturer. The hidden dangers of low cost are clearly explained. It also explains how to identify refrigerators. I am no longer afraid of being scammed by various soft advertisements."

DeepSeek is challenging Li Jiaqi and Dong Yuhui, who mainly promote products and make recommendations. When DeepSeek begins to deconstruct the "recommendation logic" of Li Jiaqi and others, the transfer of shopping power from KOL-dominated to AI-enabled may change the e-commerce ecosystem.

However, at present, "D Selection" cannot replace the shopping guide of the anchor.

The main problem is "AI hallucination". This is related to the training data of large models. The large amount of Internet data used to train generative AI is a mixture of accuracy and bias, and the model may inadvertently absorb and replicate this data.

DeepSeek is also unable to avoid the problem of AI hallucination, which also results in D selection failing to play the most important screening role, and human factors are needed to screen the AI ​​output content.

This is also one of the advantages of anchor recommendations. Although anchors cannot make accurate one-to-one recommendations, there are professional organizations behind them to select products. "D selection" cannot completely replace anchors.

For example, among the sunscreens DeepSeek recommended to Liu Xue, the first one ranked was a white-label product. From the reasoning process, we can see that the reason DeepSeek recommended this white-label product was because it captured multiple marketing advertisements. Users do not want to be disturbed by soft advertisements, but DeepSeek will use the marketing account's articles as reference sources.

However, Liu Xue, who has seen many soft ads, has her own judgment and finally made a purchase decision based on her own cognition. This does not affect DeepSeek recommendations becoming a major trend of the times and gradually entering the consumption decision-making process of young people.

In addition, unlike one-stop shopping and ordering in the live broadcast room, after Liu Xue sought help from DeepSeek, she still had to jump to the e-commerce platform to compare prices and make transactions, and low prices are precisely the core advantage of Li Jiaqi and others.

From this perspective, e-commerce platforms have more advantages in AI recommendations because they can directly attach shopping links after the recommended references to form a closed transaction loop.

02

In the past few years, the e-commerce industry has experienced a round of traffic migration towards recommendation methods.

"On this year's Double 11, the traffic brought by personalized recommendations has exceeded the traffic brought by searches," said Jiang Fan, then president of Taobao, in 2018. Pinduoduo, which was on the rise at the time, was known for its algorithmic recommendations. At the same time, the industry also derived forms such as anchor recommendations and content recommendations.

Now, e-commerce platforms are trying to seize the new form of recommendation, AI recommendation. In fact, e-commerce platforms have already entered the era of AI big models, and the shopping guide scenario formed by AI + e-commerce is not unfamiliar. However, e-commerce platforms have not given AI products a first-level traffic position.

As early as April 2023, former Alibaba CEO Zhang Yong said that all Alibaba products will be connected to the "Tongyi Qianwen" model, and the connected products naturally include Taobao.

However, Tongyi Qianwen does not appear directly in the search portal of Taobao APP, but connects to Tongyi Qianwen through Taobao Wenwen to realize AI service. Users need to search for "Taobao Wenwen" in the Taobao search box before entering the AI ​​assistant page to enter questions, including product selection guides, etc."

In 2023, JD.com also tried to integrate JD.com's Yanxi Big Model into multiple supply chain scenarios such as consumer shopping guides and merchant operations. Similarly, the "Yanxi Big Model" did not appear in the first-level entrance of the APP, but you need to search for "JD Jingyan" in the search box of the JD.com APP.

Image source: JD.com

At the end of 2024, Douyin Mall APP will launch AI shopping guide service "Smart Shopping", which is provided by Doubao Big Model. Its entrance is located in the personal center of Douyin Mall APP. This is an AI shopping guide product, and users can ask AI for consultation and comparison services for products. In terms of service experience, it is similar to Taobao Q&A and JD Jingyan.

The latest news is that the Douyin App is opening two super entrances for Doubao in internal testing. The entrances are located in the Douyin short video interface and the message list. Of course, judging from the entrance location, access to Doubao is mainly for Douyin content field services. At the end of 2023, Douyin also launched the "AI Search" service, which is supported by the Doubao large model.

Image source: Douyin embedded AI search

In addition, Kuaishou and Pinduoduo have not yet directly connected the big model into the APP.

Although the above-mentioned e-commerce platforms have already laid out AI e-commerce, judging from AI shopping guide capabilities alone, their shopping guide capabilities are slightly inferior to DeepSeek.

Ganan, the product manager of Taobao Q&A, once revealed that their original idea for Taobao Q&A was that users would not need to search on Baidu, Xiaohongshu, or Zhihu when shopping, but could complete the experience directly on Taobao.

But in actual experience, Taobao Wenwen has a word limit for questions, and questions only support text within 500 words; secondly, the content sources of Taobao Wenwen are not as many as DeepSeek. For example, for the same question "I am 27 years old and have sensitive skin. Please recommend me a sunscreen priced between 100 and 200 yuan." Taobao Wenwen's search results show that it references 6 articles; DeepSeek's search results show that it references 49 web pages.

Taobao Q&A does not support deep reasoning and deep thinking. For example, in the above question, Taobao Q&A simply gives the brand and a simple explanation of the ingredients; while DeepSeek lists the characteristics of the ingredients, applicable scenarios, precautions, and purchase suggestions.

Image source: Taobao Q&A

JD Yanxi has similar functions. Although JD Yanxi's entire product line was connected to DeepSeek in February this year, JD Yanxi did not demonstrate DeepSeek's deep thinking ability. Asking the above questions on JD Yanxi, the results are even simpler. JD Yanxi will directly attach several product links and a summary.

Douyin AI search based on the Doubao model does not provide a comprehensive and in-depth analysis of the same question like DeepSeek.

Some e-commerce platforms have also announced that they will integrate DeepSeek, but the function is not for consumers, but for merchants. In January this year, 1688, a subsidiary of Alibaba, announced that it would integrate DeepSeek.

Ziyibang found that the "1688" APP was not connected to DeepSeek, but the "1688 Merchant Edition" APP was connected to DeepSeek. In the middle of the bottom navigation bar of the "1688 Merchant Edition APP", click on AI Employee to enter the AI ​​Digital Employee page, and select "DeepSeek-R1" at the bottom to think about the problem.

Image source: 1688APP merchant version

As early as 2024, 1688 launched a free AI digital employee, which combines the capabilities of data specialists, operation specialists and other positions, and can reduce the labor costs of each merchant by an average of 4 people. Zhuo Han, the relevant person in charge, said that with the help of AI technology, buyers can place orders directly without communicating with merchants. At present, the proportion of silent orders has reached 70%, and it is expected to increase to 90% soon.

A clothing merchant who has been on 1688 for two years said that DeepSeek can currently help solve some of the questions asked by customers; but there is no real sense of cost savings yet.

Zizibang found that for the same question, the results obtained by consulting 1688 Deep Thinking as a seller are very different from those obtained by directly using DeepSeek.

After using DeepSeek-R1 to ask a question on 1688, "I sell water cups, please help me come up with a detailed promotion plan", the product gave a concise result including 6 points such as "target customer group", "optimize product information", and "participate in platform activities". It also prompted "Since there is no relevant information in the knowledge base, I will combine my own experience to provide you with a detailed promotion plan suggestion."

When the message “I sell water cups on the 1688 platform, please help me come up with a detailed promotion plan” is sent to DeepSeek, the plan it gives is very detailed and specific, down to budget allocation and risk response.

03

The core of e-commerce platforms is to facilitate transactions. On the merchant side, there is already the first e-commerce platform that has connected to DeepSeek; however, on the consumer side, e-commerce platforms have not yet directly connected to DeepSeek, and the current layout of e-commerce platforms in AI shopping guides is not satisfactory. How to capture C-end users and accurately tap into consumer needs is a difficult problem facing them.

Several Internet platforms have announced their access to DeepSeek, and it has already spread to many industries such as automobiles, energy, finance, cloud, and communications. For example, Tencent and Baidu have already brought DeepSeek to their core apps.

However, in the competition to connect to DeepSeek, except for Alibaba's B2B platform 1688, major e-commerce platforms such as Taobao, JD.com, Pinduoduo, and Douyin were collectively absent.

"Blue Whale News" also reported this phenomenon, saying that compared with Tencent's enthusiasm, Alibaba and ByteDance are more cautious and currently focus on office scenarios. Taobao, Alipay, Douyin and other core applications in their business ecosystem have not yet announced cooperation.

In response to this question, Jiang Yi, founding partner of Hengye Capital, who participated in the big model investment, first explained from the aspect of "the adaptability of Tencent and Baidu to DeepSeek". Baidu's search engine relies on big models to answer open questions, and its core indicators are directly linked to the big models. WeChat's "AI Assistant" is also highly compatible with DeepSeek's multimodal dialogue capabilities.

But e-commerce platforms have many factors to worry about.

First, the data is highly sensitive. The core data of the e-commerce platform is directly used for core business logic such as recommendation algorithms (such as collaborative filtering) and advertising (such as real-time bidding). In addition, the cost of model adaptation is high. If an external large model is connected, problems such as data desensitization, feature engineering reconstruction, and model fine-tuning need to be solved.

Secondly, e-commerce platforms (such as Taobao) have built a complete AI technology stack (such as the NLP engine "Taobao Wenwen" based on Tongyi Qianwen, and visual recognition based on SenseTime). Switching to external models requires reconstruction of the underlying architecture, and the ROI risk is high. DeepSeek currently opens capabilities in the form of APIs (such as text generation and code understanding), but e-commerce platforms need to be deeply integrated into business flows (such as search ranking and customer service conversations), involving complex system docking (such as message queues and log points).

Not to mention the integration with DeepSeek, e-commerce platforms are not even very enthusiastic about integrating with their own large model products.

So far, Taobao, JD.com, Pinduoduo and other platforms have not placed their self-developed large model products on the first-level entrance of the APP homepage, and related AI functions need to be triggered through specific search paths. Only Douyin is conducting grayscale testing on its self-developed Doubao large model and plans to integrate it into the Douyin APP.

This may be constrained by multiple factors. In terms of technical reliability, although AI recommendation systems represented by "D Select" have search engine functions, they have "AI illusions" that lead to inaccurate information, which may affect user experience and platform credibility.

From the perspective of business logic, big model recommendations are grabbing the cake of search traffic and recommended traffic such as "Guess You Like" traffic. Previously, e-commerce platforms have established a mature paid traffic system. Big model recommendations will gain higher weight, which will weaken the value of search traffic and recommended traffic to a certain extent and disrupt the original balance of the business ecology.

Of course, AI search results can also be commercialized. Before the technology maturity and commercial value are fully verified, the platform may be more inclined to prioritize low-risk scenarios for function embedding.

However, not connecting to DeepSeek does not mean that e-commerce platforms give up "deep thinking", but rather achieve it through self-developed large models.

Ali's Tongyi Qianwen official account Qwen recently released the QwQ-Max-Preview deep thinking model. Beijing News Shell Finance also reported that recently, ByteDance's AI assistant Doubao is testing different experimental versions of the deep thinking model on a small scale.

In February this year, Leifeng.com reported that Pinduoduo has set up an e-commerce recommendation model team, headed by a core member of Baidu Phoenix Nest. Sources revealed that currently, Pinduoduo's big model products are divided into several applications: price comparison system, recommendation, advertising, search, customer service, etc.

Obviously, e-commerce platforms also need to think deeply about big models.

However, compared with these products that are still on the road, DeepSeek, which has been verified on a large scale, has more advantages and can better grasp the consumption psychology of this generation of young people in terms of shopping guides.

Li Jiaqi and Dong Yuhui have used poetry and trust to build bridges with users, while DeepSeek's answer is like a precision engine that penetrates needs, scenarios, etc. one by one. In this migration of "recommendation rights", technology may solve most problems.

Jiang Yi also mentioned the technically feasible path for DeepSeek to enter the e-commerce industry. For example, cross-platform data joint modeling can be achieved through privacy computing frameworks (such as Google Federated Learning), such as providing "user preference prediction" services for e-commerce platforms without exposing raw data; developing review robots to automatically detect whether product descriptions violate advertising laws (such as banned words such as "most effective" and "national level"), and reducing platform supervision costs.

However, all major e-commerce platforms were silent in this event. E-commerce platforms also need a "Tencent" to be the first platform to access DeepSeek.

References:

"TikTok connects to Doubao, ByteDance launches counterattack" Tech Planet

"Taobao internally tests the large-scale AI application "Taobao Ask", how will e-commerce + AI play out?" First Financial Daily

"Doubao: Testing different experimental versions of the deep thinking model on a small scale" Beijing News

《Real "Open" AI is coming! DeepSeek is about to open source 5 code bases, what does this mean》Daily Economic News

"Ali, Douyin, and Kuaishou have made great efforts to develop AI" E-commerce News

"Pinduoduo AI Big Model Secret War" Wall Street Journal

"Taobao has a new entrance" - Tianxia Online Merchants