AI Taobao store, after 2 months of operation

Lumi Micro Store was founded in April 2022, and has sold 600 short-sleeved shirts and 100 skirts in more than two months. More than 90% of the store's content is completed by AI, including the store name, product title, detail page, product pictures, posters, etc., and people are just supervisors. Therefore, this article systematically explores the impact of AI on e-commerce and related industries. Through the Lumi Micro case, the advantages and limitations of AI are analyzed, and the changes and challenges that AI may bring are predicted. This provides readers with a more authoritative reference for understanding the future relationship between AI and e-commerce.

From this perspective, AI is almost the same as humans when it comes to opening online stores.

A Taobao store named "LumiWink" (hereinafter referred to as "LumiWink") was launched in April this year. Two or three people, with the help of AI tools (mainly ChatGPT and Midjourney), spent a week or two to complete the construction of the store and the listing of goods in a "formal" manner - this task would have required at least a dozen people and a considerable investment of money and time.

As of June 27, Lu Miwei has sold more than 600 short-sleeved shirts and 100 skirts, of which the short-sleeved shirts cost 69 yuan and the skirts cost 89 yuan.

It even got on the express train of the leading live broadcast organization "TVB Knowledge".

"TVB Knows Goods" brings goods to AI short-sleeved

"E-Commerce Online" interviewed the store manager (36Kr), trying to understand through this practical case: to what extent AI can empower the e-commerce industry, and which links are still difficult to replace.

And whether this will lead to the reconstruction of e-commerce production relations.

1. Copywriting, product images, and posters all rely on AI

If the store’s homepage didn’t openly state “AI produced”, I believe no one would have noticed that this is a Taobao store with an “AI concentration” of over 90%.

Store Home

"It looks no different from an ordinary Taobao clothing store, and is even more refined than ordinary small and medium-sized stores." A heavy online shopper who has browsed Lu Miwei told reporters. But she also pointed out that Taobao stores are now more personalized, with not only product posters on the homepage, but also live broadcasts, short videos, member areas and other content integration. From this perspective, as a new store, Lu Miwei's homepage is still a bit thin.

36Kr told us that almost all the pictures in the store, whether product pictures, posters, or even store logos, are generated by Midjourney (some of which involve post-processing). At the same time, ChatGPT also helps write the copywriting, from the store name, product title to the detail page.

“We are in the role of an overseer.”

The process of AI implementing commands issued by humans can be briefly expressed as: put forward your requirements for a piece of clothing to ChatGPT and let it generate the corresponding clothing parameters, such as fabric, material composition, pattern, etc. Then, based on the clothing parameters and the specific description of the clothing, a set of instructions is generated to let Midjourney create product pictures.

From copywriting to pictures, product details page with deep AI involvement

Two months after the store opened, the clothing designed by AI, a short-sleeved shirt and a skirt, has gone through the entire e-commerce process of "design-production-fulfillment", and buyers have received the goods and put them on. The monthly sales of short-sleeved shirts even reached more than 400 pieces, and there was not a single negative review among the more than 100 reviews.

In the whole process, AI's main capabilities are reflected in the front-end operations of stores such as design and live broadcasting. Lu Miwei also purchased an "AI anchor" from a digital human company. "The models, the scenes in the live broadcast room, the pictures of the displayed products, and even the 'talking skills' and keyword instructions in the real-time replies in the live broadcast room are all completed based on AI's productivity."

However, AI cannot help with production, fulfillment and after-sales service for the time being. "We estimate that AI can currently participate in 30% to 40% of the entire store operation," said 36Kr.

We observed that the click-through rate of the cover image of Lu Miwei's AI live broadcast was 1%, and the click-through rate of the product was 5%. In this regard, we consulted Wang Ming, who has been engaged in live broadcast traffic for a long time, and combined with the monitoring data of some third-party platforms, we learned that there is still an objective gap between AI live broadcast and real-person live broadcast.

"For mature stores, if the fan return rate (the proportion of fans in the live broadcast room) is below 30%, then the cover click rate of the live broadcast by real people is 5%-8%, which is relatively good. In comparison, the cover click rate of the AI ​​live broadcast room is relatively low. For excellent real anchors, the product click rate can reach 5%-20%."

E-commerce Online once reported that "299 yuan per set, 24 hours a day, AI anchors are involved in live broadcasting". Full hosting and 24 hours a day are the advantages of AI anchors. However, the lack of interactive capabilities and limited product categories are also the main problems facing this technology. 36Kr also said that for some stores that lack live broadcast infrastructure, AI anchors can reduce costs, but in terms of live broadcast effects, they are still not as good as real people.

AI anchors have shortcomings, and the same is true for product images produced by AI. In terms of lighting and model expressions, they are still inferior to real models.

"Some businesses will use AI software to put real clothes on AI models and generate model pictures, which can save model costs. But on the other hand, AI's restoration of real clothes still has flaws. Some details will be distorted, and even the clothes will be slightly different in each generated picture." An enthusiast with more experience in using AI painting added.

However, the biggest impact that AI anchors bring to the industry is not how realistic or flexible they are, but their low technical and financial thresholds, and near-zero marginal costs, which allow AI anchors to be used on a large scale. In its report on AI anchors, Guyu Lab mentioned a case: a boss bought 100 computers, rented a 1,000-square-meter office, and engaged in AI live broadcasting.

In the past, the competition in e-commerce live streaming was about products and anchors. But in the future, technology may become another important dimension.

Therefore, we believe that the biggest revelation that AI product images bring to the industry is not simply how realistic the product image is, but that it is not only drawn, but also enters the factory assembly line and becomes a real product.

2. Rejected by 20 Factories

When the store first opened, Lu Miwei designed many "product pictures", but finally chose to produce these two items. The judging criteria were relatively simple - consumers had a lot of inquiries about these two items in the background.

"The production of short-sleeved shirts went smoothly. As the color of skirts is gradient, more considerations need to be given to the craftsmanship, style and material. Therefore, many factories were unwilling to take it on at that time." 36Kr introduced that it was rejected by at least 20 factories. On the one hand, the mass production scale of new stores is relatively small. On the other hand, AI product pictures do have more room for improvement between design and production compared to traditional manual work.

AI-involved gradient skirt product image

Finally, Taobao platform contacted an experienced supplier for them and completed the “from picture to product” link.

"The gross profit is not high. We roughly calculated it to be around 20%, which is definitely lower than the average level of the clothing industry." 36Kr said that the primary consideration for opening the store this time was not the business model, but the technical feasibility. So they also wrote on the product details page: The T-shirt is priced at only 69 yuan, and including labor costs, this price is almost the cost price. Our original intention is to let more people feel the possibility of co-creation between humans and AI through this T-shirt."

This is not the first time that AI has been applied to the clothing industry.

Previously, an AI painting enthusiast on a social platform created a photo of an AI girl titled "Forest-style Black Hair". However, what became popular was not the hair color, but the "little green skirt" worn by the girl. A number of businesses took the opportunity to put this little green skirt into factory production.

At present, this single product has sold thousands of units. For small and micro businesses, this sales scale can obviously cover the production costs and obtain considerable profits.

"In fact, as long as the AI ​​product images are generated in a 'realistic' style, Chinese clothing factories are strong and can basically produce them. It's just a matter of cost." Liang Yi, a fur merchant in Haining with more than ten years of experience, told reporters that he has seen some AI-generated clothing on the Internet and it is not difficult to put them into production.

The once popular "AI little green skirt"

It is understood that whether Lumiwei will continue to launch new products on a regular basis is still uncertain. The project has been temporarily handed over to a more suitable project team.

3. Reconstruction of e-commerce production relations

The essence of the retail industry is goods and services. From this perspective, AI's participation in the e-commerce industry has not changed the essence of the industry, but it has indeed shifted the "path" of goods and services.

In the past, the traditional clothing production path was "design-production-sales". With the low-cost design capabilities and retail infrastructure (online stores) brought by AI tools, small and micro entrepreneurs without products can also turn their ideas into "pre-sale products" and place them in online stores efficiently built by AI to acquire customers in the early stage and decide whether to go into production.

Therefore, the revelation of AI to the e-commerce industry lies in that it penetrates most links of the industry and builds its own collaborative ecosystem, making repetitive operational actions, and even creative clothing design, posters, copywriting, etc., become universal resources that every practitioner can access.

In fact, e-commerce platforms are already doing this. For example, Alibaba Cloud's "Tongyi" big model, with its generative dialogue and image capabilities, is planned to be applied to e-commerce scenarios. Titles, pictures, posters, and advertising placements may all be deeply AI-based in the future.

"Systems like intelligent title generation, smart image cutting, and product scene changing are all products that already exist on e-commerce platforms. It's just that in the past, the platforms themselves did this, but now external AI products are forcing the platforms to improve," Liang Yi added.

In the past, fast fashion brand Zara was able to quickly roll out its clothing on the market based on its quick insight into popular styles on the runway each quarter and the quick response of its flexible supply chain. But in the future, fast fashion's fashion sense and trend insight capabilities at the front end may face the challenges of AI painting's large-scale production and diversified creativity.

"This may become a revolution in the definition of fashion. In the past, clothing design originated from designers. But in the future, after this design capability is popularized by AI, folk designs may emerge. In other words, everyone may become a designer. Then it comes back to the question of who the market will buy into. But fashion itself is a kind of discourse power, and the industry status and ability of excellent designers may be difficult to replace." Liang Yi said.

In addition, Yao Huanqing, associate professor at the School of Intellectual Property of Renmin University of China, pointed out that currently, the laws of various countries do not grant copyright protection to products produced by artificial intelligence. When pictures generated by artificial intelligence are commercialized, they do not involve copyright issues. "Clothing pictures generated by AI are not protected by copyright law because they are not the results of the intellectual creation of a 'natural person'."

But he also added: "There is limited human involvement in AI painting, but people connect AI-generated products together and display them on a T-shirt, which constitutes a new expression and may fall within the scope of copyright protection."

36Kr defines itself as an "overseer" in the practice of Lumi Micro Store. In the future, not only the e-commerce industry, but also the widespread penetration of AI in the industry will make more and more people become "overseers". By then, the understanding and application of AI technology may become a common requirement for practitioners in various industries.

Author: Shen Songnan

Source: WeChat public account "E-commerce Online (ID: dianshangmj)"