If competing on low prices is a form of internal competition, then the only way out is to compete even more, through technological innovation.
E-commerce giants have been struggling for a whole year over "low prices", from adjusting business to adjusting organization, and even replacing business leaders, but none of them seem to have been able to shake Pinduoduo's structural "low prices". As a result, Pinduoduo has become the "cliff of reflection" for Jack Ma and Liu Qiangdong.
Sandalwood e-commerce monitoring data shows that from October to early December, the weekly GMV growth rate of Maogou, Douyin and Kuaishou has recovered due to the low base last year. Compared with the previous quarter, it has improved slightly, from 5% to 8% year-on-year, among which Douyin has the strongest performance.
The above-mentioned institutional analysts mentioned that in terms of categories, liquor, contact lenses, outdoor sports, and health products performed well, all achieving double-digit growth, while sales of cosmetics and clothing were relatively sluggish and underperformed the overall market. CITIC Securities previously estimated that during the Double 11 period, Pinduoduo's GMV growth rate will increase by more than 20%.
Data source: Sandalwood
Faced with the aggressive Pinduoduo, Taobao, Douyin e-commerce, and even Baidu, which has been defeated repeatedly, have all turned their attention to AI.
Earlier, Jack Ma mentioned in the intranet, "AI e-commerceThe era has just begun, and there are opportunities for everyone. "Wu Yongming's concurrent appointment as CEO of Taobao and Alibaba Cloud is seen as a prelude to the promotion of AI e-commerce. He mentioned earlier that AI technology will become the core means to achieve the two goals of price power and consumer-centric service.
Not only Taobao, but even Baidu has rekindled its e-commerce passion for the fourth time.
On January 10, Baidu E-commerce announced the Baidu version of the AI e-commerce plan, which will mainly use free AI digital people, a modified search engine, and Baijiahao pictures and texts to restart the e-commerce business. At the same time, Li Yanhong once again mentioned e-commerce in his OKR for the new year.
In addition to Taobao and Baidu, JD.com and Douyin e-commerce are also eyeing the market. For example, at the end of November, Douyin established a new AI department "Flow". Before the New Year, the internal test "Worth Watching" was launched. Users can jump to the product page through the tags in the short video. This form is regarded as the next generation trailer solution.
AIE-commerce is not a new thing. Its essence is an "efficiency revolution".AIWith the big model, there is a more efficient connection between the platform, merchants, users and commodities. Its core logic is to reduce costs and increase efficiency through technical means, thus providing another solution besides Pinduoduo's structural "low price".
Digital people under the “free model”
Late at night, Acheng's (pseudonym) office was empty, with only a few computer screens on.
"In the past, if the computer performance was not good, the live broadcast with digital human would crash. Now AI technology has improved, and the computer can be used in the background to broadcast." Acheng is an online operator of a consumer product. He mentioned that technological progress has not only made live broadcasts smoother, but also begun to show cost advantages. Live broadcasts with real people cost 300-400 yuan per hour, and cartoon digital human costs about 20,000 yuan a year, which is less than 55 yuan a day.
This does not mean that digital humans can dominate the world.
At present, Ah Cheng is troubled by two problems. One is that the live broadcast of digital people needs to pre-edit the script, which has poor interactivity. The other is that platforms such as Taobao often set many restrictions on the live broadcast of digital people, including traffic. "The cost is already lower than that of real people. If the traffic is still higher than that of real people, who will use real people in the future?"
From a cost perspective, the costs of the three forms of live streaming e-commerce, namely, Dabo, store broadcast, and digital humans, are completely different.
Industry insiders have calculated the requirements for maintaining positive gross profit under different live broadcast formats. The comprehensive fees (commission + slot fee) of top and middle-level broadcasters are between 30% and 70%, and the brand ROI is usually controlled between 1% and 3.3%. The advantage of broadcasters is that they can quickly increase the volume.
It is precisely because of Dabo’s gross profit requirementsVery highTherefore, for old brands that are very mature in the Taobao system, their gross profit margin limits their switch to Douyin.
"Big brands mainly broadcast live on Taobao, and the maximum frequency of live broadcasts is basically 20 times per month." The above-mentioned person mentioned that the situation on Douyin is completely different. Although brand merchants have store broadcasts on Douyin, they still have to rely on large-scale live broadcasts to increase sales. "As far as I know, many new brands can broadcast more than 20,000 times on Douyin in a single month. This exposure is unparalleled."
Perhaps seeing the trend of old brands being "tired of Douyin", Taobao has continuously strengthened brand store broadcasts in the past year. According to estimates by two brands, the cost of live store broadcasts is at least 400 yuan/hour/person. If there are multiple anchors + supporting actors + field control, as well as the amortization of the scenery + equipment, the cost will be higher. "The problem with store self-broadcasting is that the cost is rigid. Compared with Dabo, it is more like a heavy asset investment. If sales do not increase, it means losing money."
In order to solve the rigid cost problem of live-action store broadcasts and provide all-weather coverage, many brands, especially many fast-moving consumer goods and health care brands, have introduced digital humans to form a model of live-action store broadcasts + AI self-broadcasts. A relatively well-made digital human costs around 100,000 yuan a year, and the cost of the "standard product" provided by the platform/service provider - a cartoon digital humanlowestIf we can achieve 20,000, the cost can even continue to drop.
Perhaps in order to seize the minds of users, large companies such as Baidu and Taobao, as well as small companies such as Silicon Intelligence, have used the "free strategy" to attract customers in the past year. Digital people are actually a low-cost "online shopping guide", which is, in a sense, a substitute for live broadcasts of real people.
In comparison, JD.com has not set such a high tone, using digital people as a supplement to the live broadcasting format. Considering the efforts to promote live broadcasting during the Double 11 period, the priority of digital people has further declined. Earlier, Shang Xin, vice president of JD.com Group and head of JD.com Retail Technology R&D and Data Center, had clearly mentioned that in the short term, it mainly exists to fill the gap in live broadcasting.
Earlier, JD.com’s technology department believed that the combination of digital humans and large models was more suitable for implementation in customer service scenarios. One reason may be that JD.com has a stronger need to reduce costs.
In addition, the operation of digital people varies in different industries. For example, categories such as food and beverages, toiletries, and health care products have a higher level of awareness, which may be closely related to the intensity of industry competition and the industry's comprehensive online gross profit.
Data source: Sandalwood
According to Sandalwood e-commerce monitoring data, in OctoberFirstSince the beginning of the week, food and beverage, personal care and household cleaning, and health care products have all rebounded to a certain extent. However, from the growth situation, it can be seen that the growth rate during the promotion period is not as fast as that during the non-promotion period, which shows that the impact of daily operating investment on the growth rate is more significant.
The open paradox of intelligent customer service
JD Shangxin talked about the integration of AI technology and e-commerce, focusing on four areas: smart shopping guides and customer service, smart quality inspection, R&D efficiency improvement, and operation and maintenance.
Shang Xin’s statement reflects the commonality of comprehensive e-commerce companies in carrying out AI transformation.
In December last year, Taotian merged more than 20 teams responsible for AI into four, namely Alimama, C-end consumers, B-end merchants, and industry-specific applications.
Specific implementation scenarios include marketing represented by digital humans and search recommendation optimization, and services for C-end consumers such as Taobao Q&A and intelligent customer service; the third is intelligent business tools for merchants, such as the use of multimodal generation of picture and video detail pages. If the technical services and machine deployment provided by Alibaba Cloud for Taobao are added, its logic is no different from that of JD Retail.
Intelligent customer service is currently the most mature application scenario of AI e-commerce.
Chen Zhe, VP of Technology at Wisdom Tooth Technology, told Photon Planet that intelligent customer service is mainly divided into two implementation forms. One is a product that extracts knowledge based on the knowledge base and generates it using AIGC capabilities. "The advantage of this solution is that it can reduce manual maintenance costs and increase the rate of answering questions; the disadvantage is that because the robot automatically learns and uses a large model to answer, the answer is not so rigorous."
Disadvantages hinder landing.
"When Wenxin Yiyan or Xinghuo produces a wrong answer, users will just laugh it off, but in serious business practice, a mistake means immeasurable losses." The comparison made by Cao Peng, chairman of JD Group's Technology Committee and president of JD Cloud Business Unit, expresses that many companies will be more cautious in the application of technology.
Another way of integration is to allow customers to upload documents, automatically generate them through large models, and build a knowledge base. Generally speaking, it only needs to be approved by auditors before it can be put online.
From a business model perspective, the extent to which intelligent customer service can reduce costs and increase efficiency determines the profit margins of related service providers. According to calculations by Wisdom Tooth Technology based on dozens of retail customers, after the addition of the big model, the accuracy of the original answer rate has increased from 40% to 70%, and the cost rate is only 20%.
According to Chen Zhe, retail brands are more willing to purchase this kind of service. "Before 2023, we may not be able to convince customers even if we popularize it. Since 2023, many more customers have come to us."
Offline retailers prefer third-party smart customer service products for multiple reasons, including more obvious cost reduction, offline shopping guides and private domain services. In addition, offline retail is not like online, which is dominated by platform economy.
In fact, current e-commerce platforms are not friendly to third-party service providers, and most platforms do not open API interfaces to the outside world. According to an industry insider, the sensitivity of private domains and data makes platforms unwilling to open up to third parties.
The fact that the platform does not open its interfaces does not mean that the results of large companies' self-developed products will be very good. On the contrary, under the pressure of KPIs such as revenue and conversion, e-commerce platforms are somewhat uninterested in investing in R&D of universal functions such as intelligent customer service.
A service provider said that intelligent customer service needs to effectively identify consumer problems, which depends on two points: whether the pre-buried script pool is rich and whether dynamic adjustments are timely. "Taotian is currently only in a semi-AI state. AI identifies general user needs, and those more in-depth problems are solved by transferring to humans."
"Flower fists and embroidered legs"?
In addition to digital humans and intelligent customer service, AI and big models also have broader application prospects in marketing, but in the short term they are limited by technical bottlenecks and organizational systems.
"After the organizational structure adjustment, Taotian's technical line has not made any moves yet." An insider believes that Taotian needs to solve the problem of separation between technology and operations in order to do AI e-commerce. "The internal assessment of a new function is either the user time or the user call rate, but there is no assessment requirement for whether there is a conversion or how to convert the technology."
In addition, Taotian still has a lot of room for improvement in its recommendation algorithm. A simple example is that after a user has placed an order to purchase a product, similar products will continue to be recommended in the following period of time - this shows that Taotian's recommendation algorithm still does not take historical orders into account.
This may be limited by computing resources. Earlier, when JD.com was testing the big model in e-commerce, it also tried to include users' historical orders in the shopping guide. However, the large amount of data and exponentially growing calculations made this function expensive and inefficient.
This is the case even when incorporating historical orders, let alone solving multimodal problems, which allows merchants to completely solve the problems of product materials and content operations.
There are thousands of paths for AI e-commerce, but the core of all of them is "reducing costs and increasing efficiency." Only when costs and efficiency can help merchants release more gross profit margins can comprehensive e-commerce platforms have a chance to compete with Pinduoduo's "low prices."
"We have calculated before that under the same conditions, the basic fees + marketing fees of different platforms are actually similar. In other words, when merchants decide where to go, they still consider the business environment more."
An industry insider believes that in addition to "low prices", Pinduoduo also has the advantage of low barriers to entry for merchants.maximumThe problem is the complicated price stacking, which ones can be stacked, which ones are parallel stacking, whether to set membership prices, and even having to solve a two-variable linear equation. "
Over the years, pricing strategies have continued to stack up, and brands often encounter operational accidents during big promotions and are fleeced. For example, the Guo Xiaoyun Taobao flagship store in 2019, the JD Changhong Kitchen flagship store in 2020, the Yuanqi Forest Tmall flagship store in 2021, and the Minitutu Tmall flagship store in 2023. Almost every year, there are cases where merchants suffer huge losses due to price setting errors.
The problem facing cats and dogs may not be difficult. To achieve low prices, we still need to solve the business problems of merchants. Wang Xing once learned a truth from an old predecessor: there is only one way to succeed in retail, and smart people waste their energy. The so-called AI e-commerce now seems to be a reverse engineering.