Text: Yicai Business School, Chen Xianfeng
Editor|Wu Lingwei
2023 is the first year of AI. ChatGPT, an artificial intelligence software, became an instant hit, becoming the fastest-growing consumer app in history, and a keyword throughout the year.
GPT is listed in Nature's 2023 list of people of the year — it's not human, but it's revolutionizing the world. According to Collins Dictionary, people will mention "AI" four times more times in 2023 than last year.
On July 7 alone, Huawei, Alibaba, Tencent, SenseTime, and JD.com released large models on the same day. The last time it was so lively was when blockchain and the metaverse exploded – both of which proved to be short-lived, and only AI was seen as better embedded in the business of companies, as a new ticket to the next era.
ByteDance, Alibaba, JD.com, Xiaohongshu, Kuaishou...... Internet giants have invariably chosen AI as a breakthrough, giving rise to a series of innovative application scenarios, which not only reshape the shopping experience of consumers, but also silently change the competitive landscape of the e-commerce industry.
Yicai Business School will depict the silhouette of the AI e-commerce era from two aspects: the AI layout of major e-commerce companies and the AI application of enterprises.
01 Which e-commerce platform has moved on AIGC?
AIGC has become the "number one project" for large Chinese companies. The basic idea of doing AI is the same as OpenAI: one hand to make a large model, the other hand to do the application, the former to provide technical foundation and capability support, and the latter to continue to provide user feedback and data. Therefore, we will judge the investment of each e-commerce platform in the AIGC field from two dimensions: first, the organizational responsiveness, and second, the completeness of the layout.
After comparing the AIGC large models and application products released by Taotian, JD.com, Douyin, Kuaishou and Xiaohongshu, we found that in terms of organizational responsiveness, Alibaba launched the largest model products the fastest;
1. Organizational responsiveness: Ali is the earliest
From the perspective of the launch time of large-scale model products, Alibaba is the earliest. In April, Alibaba Cloud released the large-scale model product "Tongyi Qianwen" and announced that all of Alibaba's products will be integrated with the "Tongyi Qianwen" large model. In December, Taotian Group released the large model "Xingchen" (formerly known as "Turing"). In July, JD Cloud released the Yanxi model for vertical industries. In August, Kuaishou released the self-developed language model "Kuaiyi", and in the following September, it released the self-developed Wensheng graph model "Ketu" again. ByteDance, which is considered to be not weak in technology and resources, only launched an AI dialogue assistant based on the lark model in August. Pinduoduo is the latest, and it was only announced in December that it has formed an AI large model team of dozens of people, which will explore the application of AI large models in customer service, dialogue and other scenarios.
2. Layout integrity: Taotian has both BC applications, covering the most business scenarios
All e-commerce platforms have launched their own large models and related application tools. The first few tools they have landed are all combining their own "advantages": taking Taotian as an example, at the model level, its internally open "Xingchen" large model products are mainly used in two businesses, one is search, advertising, recommendation, and the other is the content of shopping. At the application level, toC products include "Taobao Ask", which provides consumers with shopping suggestions, and toB has more than 10 AI tools, including Wanxiangtai Unbounded Edition (to help merchants analyze data, identify target users, and intelligently generate advertising content) and Wanxiang Lab (to match models and scenes for products, and produce product images in batches), respectively, to make merchants operate more efficiently in scenarios such as intelligent delivery and product details, and JD.com starts with intelligent customer service. Its early industrial AI strategy can be summarized as "one platform, three levels of rockets": one platform refers to the Yanxi platform, and the three levels of rockets are intelligent interactive media such as intelligent customer service, intelligent marketing, and digital humans.
Content-based platforms such as Douyin, Xiaohongshu and Kuaishou are currently launching AI products that focus more on content creation and advertising materials, as well as user interaction experience. Douyin's instant creation can help merchants generate live broadcast backgrounds, short videos/live scripts, videos, etc. Cheng Yixiao, co-founder of Kuaishou, once said that the application scenario where AI is most likely to land first is in the field of short videos, and it is also the direction that Kuaishou is working hard to promote. In addition, Kuaishou's "AI Play Reviews" are used to generate creative text and pictures in the comment area, which is obviously to lower the comment threshold and improve the fun of interaction. Xiaohongshu's AI smart note-taking assistant and chat function "DaVinci" start from another angle, using intelligent technology to help merchants improve the efficiency of content production, while providing users with richer and more personalized information services.
In a word: the e-commerce platform cuts into the business first, and the content platform lands the short video first.
If you want to compare the completeness of the application level, Taotian is the most complete of them, covering content production, intelligent delivery, customer service, business data reports and other links. Content platforms, on the other hand, often only cover the content production process of merchants, and do not involve more business parts. Although JD.com also covers the customer service and marketing links of merchants, it claims to be cutting into "industrial AI" and has not launched C-end products.
02 Digital human anchor: It is the icing on the cake for big brands, and it is difficult for small enterprises
For enterprises, reducing costs and increasing efficiency is always the main theme, and the boom of digital human live broadcast stems from such a demand - it has become the first large-scale application track in the AIGC field. Data shows that in 2023, the global virtual anchor market will increase by 400% year-on-year, about more than 53 billion yuan, and more than 280,000 digital human enterprises were born in China.
However, due to the platform's rejection of low-quality digital human live broadcasts, and the cost of merchants actually operating digital human anchors, everyone found that digital humans are not as beautiful as imagined. A one-year-old digital human company told Yicai Business School that most of the small and medium-sized enterprises that have tasted digital human anchors in 2023 have chosen to retreat halfway.
Enterprises that try to build a live broadcast team around a digital human anchor often calculate such an account:
A team of live anchors, the smallest unit is 3 people: an anchor, an operation and central control - based on the employment cost in Hangzhou, the company has to pay more than 40,000 yuan per month for the team - the annual salary alone has to spend at least 500,000 yuan.
The digital human live broadcast team generally only needs 2 people: a digital human anchor, plus an operation or copywriter. According to the charging standard of the general digital human company, it is divided into two parts, one is the production of digital human image, the cost is 4000-10000 yuan, and the other is the monthly fee of 3000-5500 yuan/month. Even if the top grid is calculated, the salary cost is less than 200,000 yuan a year.
Judging from the effect, the digital human company said that in the cases it has served in the past, the GPM (turnover per thousand impressions) of the digital human live broadcast team can reach 70, and it can reach 90 at the best, even exceeding the performance of the real anchor team - note, he emphasizes the "team".
The crux of the problem lies here: the GPM performance of the digital human live broadcast team is firmly linked to the operational capability. According to the special view, the factors that determine the level of GPM include speech, scene and product selection, etc., and the test is the preliminary preparation of operation. Digital humans only replace the role of anchors, and cannot completely replace a live broadcast team.
Therefore, those companies that hope to spend a small amount of money can be a hands-off shopkeeper, and soon find that digital human anchors are not useful.
It is often the big brands that will continue to renew their fees, and their needs are more compatible with the current digital human capabilities - most of the big brands have established a relatively mature live broadcast operation team, and many brands have also launched a 24-hour "sun never sets" store broadcast. Digital humans can take over the live broadcast room during non-prime time to achieve the greatest common divisor of live broadcast duration and employment cost.
03 Acceptance spectrum of the platform: jitter
The most direct dilemma is that Douyin, which carries the most digital people, has begun to "manage" digital people. In May 2023, Douyin released China's first platform specification for AIGC, "Douyin's Platform Specification and Industry Initiative on AI-Generated Content (hereinafter referred to as the Digital Human Industry Initiative)", the main contents of which are as follows:
1. The content of AIGC should be prominently marked;
2. It is forbidden to publish illegal content, and shall not infringe on the portrait rights and intellectual property rights of third parties;
3. The publisher is responsible for the results of the generated content;
4. False propaganda is prohibited, and rumors are not allowed.
5. Real-name authentication is required for publishers.
Judging from these rules, Douyin did not kill all digital people with one shot, but only made a few specifications. However, during this period, several types of live broadcast rooms using digital humans were banned - the first category, 2D digital humans who can only broadcast mechanically and repeatedly, and the second category, the live broadcast room that bought a "public model", the same face appeared in multiple live broadcast rooms at the same time, and it was easy to be recognized by the platform as a recording and broadcast and was judged to be in violation.
Douyin's restrictions on digital human-generated content are well understood - the emotions and high-quality content required by interest e-commerce cannot be given by digital humans. Although the realism of digital humans is close to that of a real person, the interactive type is still weak. A head digital human company interviewed by Yicai Business School can not only achieve a high degree of synchronization between the digital human picture and mouth shape, but also synchronously present more detailed head movements and gestures, and even simulate the action of a real anchor trying the lipstick number with the back of his hand. However, when they face consumers' questions in the comment area, they can only identify the keyword trigger and give a prepared answer, such as the free shipping area, the cost, and the explanation of the corresponding product link.
In addition, Zheng Yanyun, an expert at the think tank of Yicai Business School, said that the crushing learning ability of AI digital humans is a shortcut for creators to chase traffic - find popular videos, extract copy script copywriting, and generate new audio and video, which is the "crushing creation" of digital human live broadcast or short video creation, but it also squeezes out the traffic space of other content creators.
"Platforms and creators should think about the value of digital human content - is it objective rational value, or emotional and emotional value? Zheng Yanyun said.
Taobao, Jingdong, Kuaishou accept
Except for Douyin, none of the other major platforms has proposed corresponding management specifications for digital humans, and they also use digital human anchors to show the diversity of the live broadcast ecology. For example, Taobao will use Lie'er Baby as the benchmark in July 2023 to realize the launch of AI head anchors for the first time. Digital human anchors have appeared in the official accounts of many Taobao factories, Taobao official accounts such as Alibaba's self-operated electrical appliance business Meow Suda, as well as Taobao live broadcast rooms of beauty and food brands such as Maybelline, Colorkey, and Xuanma. In addition, Taobao separately listed digital human live broadcast as a category, and put forward clear business qualifications and ability requirements for settled merchants.
In the 618 battle report, JD.com claimed that the number of merchants who launched the broadcast of Yanxi Digital Human has increased significantly, and the turnover achieved by the merchants has increased by 246% compared with Double 11 in 2022, and the cost is less than 1/10 of the real anchor.
These platforms show a greater degree of tolerance for digital human live broadcasts, firstly, because the platforms need to show their respective technical strengths and personally provide digital human technical infrastructure and solutions to increase additional service revenue. For example, Taobao digital human technology comes from the machine intelligence laboratory of the Damo Academy, Jingdong digital human is based on the Yanxi large model, and Kuaishou AI&User Growth Department introduced the Kuaishou AIGC digital human solution at the Kuaishou Photosynthetic Conference in August-Kuaishou Intelligent Broadcasting;
Video account, Xiaohongshu neutral
Channels and Xiaohongshu have not yet launched any regulations on digital human content, but Xiaohongshu will mark "suspected AI content generation" below the relevant posts.
The relevant person in charge of the video account also said in an interview that as long as the content does not violate the rules, digital people are not encouraged or restricted. In its live streaming promotion specifications, it is stipulated that if influencers and merchants do not show their faces for a long time during the live streaming process, affecting the viewing and interactive experience of users in the video account window, it is a violation, and Tencent will not be limited to taking measures such as limiting traffic, restricting live streaming, blocking windows, and restricting transactions.
Local life embraces digital humans
Compared with e-commerce platforms and content platforms, local life is more receptive to digital people, especially platforms such as Meituan and Fliggy, as well as Kuaishou local life business, which currently has a large number of digital human live broadcast rooms. Yicai Business School found that in the live broadcast rooms of Meituan's food group buying, hotels and homestays, the main station of Meituan's official account "Foodie Welfare Society" and its sub-stations in Shijiazhuang, Nanjing, Jinan, Jinjiang and other sub-stations, as well as chain catering brands such as Fish You Together, are all using digital human live broadcasts, and the live broadcast rooms of travel agents such as "Live and Play to Sanya" and "Tuan Tuan Go to Travel" even use the same digital human image. Except for a few brands such as Xijie Fried Skewers, KFC, and Cudi Coffee, which use live broadcasts, from Meituan's official accounts, agents to merchants, they are all using a large number of digital humans or video recordings to sell goods.
On the one hand, the local life platform is so enthusiastic about embracing digital humans, on the one hand, out of strategic needs. For example, in order to cope with the rapid attack of Douyin, Meituan regards live broadcast as a strategic focus, but most of its main merchants are offline merchants such as wine and catering that lack live broadcast capabilities. On the other hand, most of the local lifestyle products are more introductory standard products, such as hotel coupons, tour groups, meal coupons, etc., and the oral broadcast of product information can be greatly simplified into the use time, scope of use and price.
04 It's easy to buy a digital human anchor, but it's difficult to operate, how to make good use of a digital human?
1. Standard brand store broadcast. At present, digital human live broadcast is more suitable for standard products with strong brand attributes such as beauty, food, digital products, and automobiles. For non-standard products such as clothing, the cost of explaining fabrics and styles is too high, and it also emphasizes strong interaction in the live broadcast room such as try-on, and digital people cannot do it for them.
2. Select the corresponding platform according to the platform's tolerance for digital humans. The platforms present a spectrum of acceptance, arranged from small to large, in order:
Tik Tok
The platform's attitude towards digital humans depends on three points: first, the urgency of content supply, second, the dependence on high-quality content, and third, the platform's preference for commodity content or entertainment content.
For example, Douyin, a platform that relies on high-quality content, needs to encourage creators to continue to produce good content through mechanisms and products to continue to build a time black hole. This is obviously not possible with the current digital human short videos and live broadcasts. For example, Taobao JD.com has a demand for merchants to increase the duration of store broadcasts, and in turn, it also needs to accept the processing methods of merchants to reduce costs and increase efficiency, and propose corresponding solutions;
3. Two modes to meet the requirements of "real person driven" and "real-time interaction" of the platform. At present, there are two mainstream digital human live streaming methods on the market:
"The Man in the Middle" mode of intelligent takeover: By capturing the oral broadcast and movements of real people, the live broadcast room can generate real-time footage. Digital human companies with relatively leading technical capabilities, such as silicon-based intelligence and special view, have basically launched a real-time synthesis drive + intelligent takeover mode: when the live broadcast room is in the non-prime time when the audience is relatively small, the digital person will take over the real person to explain the goods;
Digital and real partner live broadcast: Digital human serves as an assistant broadcaster.
Technology is always overvalued in the short term, but undervalued in the long term. From AI customer service to AI digital humans, every technology was initially regarded as a business miracle, but when we return to rational thinking, we will find that the "steps" taken by AI in the field of e-commerce are not as fast as imagined. The application of AIGC in e-commerce, from now on, looks like a child with a smart mind but clumsy limbs.
But what we underestimate is the potential of AIGC as a technology - unlike the abstract metaverse and digital currency, it is indeed a real technology written by humans line by line of code and trained by a large amount of computing power day by day. We acknowledge that the era of AI e-commerce has just begun, which shows that we are optimistic about the long term.