Alibaba Launches 'Huiwa', Will AIGC in E-commerce Marketing Become the Next Big Trend?
-
"The era of AI e-commerce has just begun, presenting both opportunities and challenges for everyone," said Jack Ma in an internal Alibaba post last November when Pinduoduo's market value was about to catch up with Alibaba. Undoubtedly, after the founder's statement, AI e-commerce will become a key focus for Alibaba in the future. Now, Alibaba's AI e-commerce has taken a substantial step forward with the emergence of a product called 'Huiwa'. It is reported that this AI e-commerce product is currently in the testing phase, and Alibaba has invited a group of Taobao merchants to trial it.
According to the official website, 'Huiwa' primarily targets Taobao and Tmall merchants and influencers, significantly enhancing their creative efficiency through features like generating marketing copy and training exclusive AI models. This aims to help products gain faster consumer discovery and preference. Alibaba stated that through 'Huiwa', they hope to create a more efficient and intelligent e-commerce ecosystem for merchants and influencers, driving continuous innovation and development in the e-commerce industry.
In fact, 'Huiwa's' selling point is quite straightforward: "Get your products noticed and loved immediately." The concept of 'getting noticed' has undoubtedly been one of the hottest trends in the e-commerce sector in recent years. As traffic dividends become a thing of the past, traffic costs have become a major concern for every e-commerce seller, while 'getting noticed' offers a low-cost traffic source akin to 'word of mouth.' Since 2021, a slew of e-commerce platforms, including Taobao's 'Guangguang,' Meituan's 'Zhenxiang,' Tencent's 'Penguin Huimai,' JD.com's 'Zhongcao Xiu,' Pinduoduo's 'Pin Xiaoquan,' and Douyin's 'Kesong,' have crowded into this 'getting noticed' trend. This time, "Huiwa" directly tackles merchants' pain points in content creation by providing two main features: text-to-image generation and AI writing. The text-to-image function helps sellers create AI models, eliminating the need to communicate with models, photographers, and makeup artists. Merchants can now use AI models to showcase their products themselves. The AI writing feature claims to generate viral marketing copy. Merchants only need to input a Taobao link, product ID, and describe the main selling points and persona, and the tool will automatically create promotional content.
In short, "Huiwa" is a cost-reducing and efficiency-boosting tool for e-commerce sellers. However, if merchants consider it a lifesaver, they might be sorely mistaken.
Let's first discuss the AI model aspect. As early as 2022, with the emergence of Midjourney and Stable Diffusion, some merchants realized that such AI image-generation tools could be used to create "digital humans." Today, many e-commerce businesses use AI models, but in practice, these models still reveal significant shortcomings. When it comes to using AI models, the specific product categories are mainly clothing, shoes, and hats, with one key technology being 'clothing change.' Currently, merchants commonly adopt two main approaches. The first involves taking photos of the clothing from various angles and then using the LoRa model for training to ultimately generate various poses of the AI model wearing the clothes. The second method is redrawing, which requires first placing the clothes on a plaster mannequin and taking photos to fix the clothing part, then using AI tools to replace the mannequin with an AI model and create the background.
The current limitation of AI models is not that they still require manual screening after generation, but rather that the effect of the clothing on the AI model is difficult to replicate in reality. Indeed, among the various AI models, only OpenAI's Sora dares to claim itself as a 'world simulator.' Currently, AI image generation tools on the market do not truly understand the laws of the physical world.
For example, during last year's Double 11 shopping festival, the 'first batch of victims of Chinese-style sweaters' that made it to the hot search was a typical case of AI model failure. Consumers complained that while the sweaters looked light and elegant on the AI models, they appeared bulky and unflattering when worn by themselves. The stark contrast between the actual product and the AI model images on the online store clearly falls into the category of false advertising. If AI-generated images have tangibly reduced product photography costs for e-commerce merchants, then AI-generated viral marketing copy remains largely a "pipe dream." Although GPT-4 and Wenxin Yiyan 4.0 have been available for quite some time, has anyone actually seen a single viral article generated by them? Even CNET, the overseas tech media outlet that pioneered large-scale AI writing adoption, has been outright removed from Wikipedia's list of "reliable sources."
The fundamental reason AI writing struggles with persuasive product recommendation articles is that current AI systems are not AGI (Artificial General Intelligence). ChatGPT, GPT-4, Wenxin Yiyan, and even the newly unveiled "world's most powerful model" Claude3 may indeed transform many aspects of human society's production paradigms. However, as OpenAI's reports have explained, AI excels particularly at repetitive and rule-based tasks that typically don't require creative thinking but rather follow fixed rules and processes. This explains why AI customer service often receives criticism - after all, customer service faces highly individualized scenarios where different consumers' inquiries can vary dramatically. In contrast, AI models are generated through a fixed process, where stable output can be achieved via a set of SOP (Standard Operating Procedures). While AI writing is feasible, AI-generated recommendations prove challenging.
Those who have used ChatGPT or Wenxin Yiyan to generate articles may have noticed that these tools excel at writing academic papers or official documents—essentially, they are a boon for standardized writing. However, they fall short when it comes to creating recommendation content for social platforms like Weibo, Douyin, or Xiaohongshu, where personalization is key. It's important to understand that the effectiveness of recommendations in driving consumption is based on the "consumption imitation" theory. Influencers' recommendation content essentially aims to inspire fans to follow their lead through exemplary behavior, while fans derive pleasure from emulating their actions. The underlying message of such content is: "Buy this, and you'll be just like me." Homogeneity is undoubtedly the bane of product recommendations, as standardized endorsements are essentially just hard advertisements. Ironically, content with strong personal characteristics is precisely what AI struggles with. Even if AI can indeed mass-produce recommendation articles, they are clearly far from being viral hits. Using AI to enhance content production in the e-commerce sector, at least for now, is merely 'seemingly promising.' However, for ordinary sellers who simply need content to fill their online stores, this level of AI is already sufficient.
Although Alibaba's exploration of AI in e-commerce is necessary, the current state of related technology is not yet capable of disrupting the existing landscape of the e-commerce industry.