Recently, AI e-commerce has been attracting increasing attention.

In June, two AI e-commerce companies secured financing in succession, with Daydream raising $50 million in seed funding and Constructor completing a $25 million Series B round, valuing the company at $550 million.

Why is AI e-commerce becoming increasingly popular? There are two main reasons:

First, e-commerce itself is a sufficiently large and continuously innovating sector. According to Goldman Sachs data, global e-commerce sales reached $3.6 trillion in 2023 and are expected to grow by 8% year-on-year in 2024. Simply put, if AI can improve the industry's efficiency by 1%, with 20% of the benefits going to these AI companies, it represents a market worth $7.2 billion.

Second, as AI is increasingly applied in the e-commerce field, the implementation path is gradually becoming clearer. On the consumer side, with the influx of a large amount of content, the value of traditional search for shopping is diminishing, and AI has the opportunity to create a more efficient matching mechanism. On the merchant side, AI helps merchants increase efficiency and reduce costs through text and image generation.

Taking advantage of this financing opportunity for AI e-commerce companies, let's also take stock of some of the trends in AI's implementation in the e-commerce field.

/ 01 / AI Shopping Guides, the Core Track for AI E-commerce Startups

Personalized recommendations are the mainstream direction for AI e-commerce startups. This time, the funded Daydream and Constructor, as well as previous companies like True Fit and Remark, have all approached from this angle.

With an increasing number of brands and channels, coupled with various recommendation algorithms and content piles, it has become increasingly difficult for consumers to find the products they want, and traditional keyword search methods are also less effective.

Fortunately, AI excels at information understanding and retrieval, so many entrepreneurs are trying to use AI to create a more efficient matching mechanism.

What Daydream and Constructor do is enable users to search for products using natural language and image recognition. Generally speaking, when you go on vacation and search on a traditional e-commerce platform, you might use specific search keywords like "sun umbrella, swimsuit, sunscreen shirt." This is a precise and targeted search pattern.

When you search with Daydream and Constructor, you don't need to enter keywords. Just describe the usage scenario to find suitable products. For example, if you're going on a vacation, you can say, "I'm taking my 3-year-old son to the beach for a vacation, and we need some parent-child style sunscreen vacation gear."

Not only that, Daydream also provides command filters. For example, if you like a piece of clothing but want it in blue, you can enter, "I like this dress, but I want one in blue," and then similarly styled blue dresses will be filtered out.

Under this search mechanism, the search results will be closer to user needs. Moreover, these search requests reflect user preferences, and as details increase and recognition efficiency improves, the personalization of search results will also increase.

Users also don't have to worry about advertisements influencing their choices. Because AI e-commerce platforms like Daydream do not participate in order fulfillment. In other words, Daydream only serves as a discovery layer for shopping, and its revenue comes entirely from commission fees, without any advertising costs.

At least for now, market data feedback has verified the effectiveness of this product. Constructor has an AI shopping assistant ASA, similar in function to Daydream.

Since its launch less than a year ago, clients (including large grocery chains, clothing brands, and comprehensive retailers) have seen an increase in website revenue of 10%, search conversion rate of 6%, and click-through rate of 7%. In the past six months, shoppers have interacted with the Constructor platform over 100 billion times, and the average customer retention rate has remained at 98.5% over the past three years.

In addition to optimizing the search process, others have also tried to establish matching mechanisms using different methods. For example, Remark's strategy is to introduce product experts.

Remark has 50,000 experts, each specializing in their own fields, including musicians, stylists, golfers, ski coaches, etc., to provide users with high-quality product consultation and discussion services.

In addition, Remark has also trained AI to assist, simulating the style of human experts to answer questions and provide timely and professional shopping guidance to users. According to the company's disclosure, clients working with Remark have seen a 9% increase in revenue and a 30% increase in conversion rates.

Overall, the essence of personalized recommendations is AI shopping guides, as most people find it difficult to accurately and completely express their needs when shopping. Having a robot that understands natural language can gradually guide users' vague descriptions into specific requirements, which is conducive to improving matching efficiency.

/ 02 / AI Is Reshaping the E-commerce Process

Personalized recommendations have enhanced the consumer shopping experience with AI. On the merchant side, AI applications are mainly reflected in improving efficiency and reducing costs.

Improving efficiency involves replacing human labor in existing processes, with relatively mature application scenarios including copywriting, review analysis, SEO optimization, and more.

In copywriting, AI can learn from previous product detail pages and combine the highlights of new products to help merchants rewrite copy. In addition, AI can also complete the writing of emails and text messages for regular communication with users.

There are many AI writing tools available now, with Jasper being one of the more well-known products. It provides templates for various types such as marketing, blogging, business, and Google. By leveraging accumulated high-quality copy to fine-tune the GPT model, Jasper prompts users to enter corresponding information in the template, and then generates high-quality paragraphs on the right side based on the template.

Last year, Amazon also launched a tool for generating product content. After merchants enter a few product description keywords or sentences, Amazon will list the content that merchants may need, such as product names and overviews. Merchants can directly adopt or continue to optimize the list's generation effect.

Moving on to review analysis, products in stores often have many reviews, highlighting strengths and weaknesses. AI can help merchants conduct intelligent analysis and provide analysis reports, such as users wanting more colors or longer charging times. This information helps merchants adjust their product structure.

Then there's SEO optimization. If merchants have independent websites, they need to write some soft promotional articles to optimize their website's SEO. In the past, it might cost hundreds of dollars to hire someone to write for a day to produce one article. But now, after merchants feed in the corresponding corpus, AI can generate an article in just one minute, and even capture category keywords to optimize the content.

After discussing efficiency enhancement, let's talk about cost reduction.

Cost reduction involves replacing most of the tasks that sellers previously needed to outsource directly with AI, with the most obvious area being image production.

E-commerce practitioners often do not possess professional image processing skills but consume a large amount of image material. For example, many sellers currently re-edit the background of manufacturers' product images, but the problem is that upstream product images may be rough and unattractive.

Now with AI image generation products, these troubles are gone. In this regard, Photoroom is the best-performing product abroad.

Photoroom first started with background editing, allowing e-commerce sellers to complete background editing (erasing, blurring, replacing, etc.) of product photos with just a few clicks, creating higher-quality product promotion images at a lower cost. Now, Photoroom can also use prompt-based image generation to quickly produce the backgrounds users want and has made many optimizations in product effects.

In China, Meitu Xiuxiu also offers an AI one-click cutout and AI background replacement function. Merchants only need to upload a product, and AI can directly recognize it, not only replacing the background but also changing the product's color.

In addition to background images, e-commerce often uses creative images or model displays for product presentation. Previously, this kind of imagery was quite cumbersome. If it involves real people, it requires the design team to conceive in advance and then invite suitable models and photographers for shooting. For the clothing industry, it takes nearly a day to shoot ten sets of clothes, with an average cost of around 100-500 yuan per image.

Now, whether it's creative image generation or model display images, AI handles it all.

For example, Linkfox abroad can automatically generate product images using AI technology. Users upload one or multiple product images, and AI can easily complete the synthesis of the product images.

Linkfox also supports uploading only a photo of a piece of clothing and then selecting an AI model to generate a product image of the model wearing the clothing. Moreover, you can adjust the model's facial expressions to make the generated model product image look more natural.

Last year, Mogujie also launched the AI commercial photography tool WeShop, which supports the generation of mannequin photos, real-person photos, product photos, toy photos, and children's clothing photos. The operation is very simple, requiring only selecting the desired image type, uploading the image, setting other requirements like text, and generating the image to complete the experience.

/ 03 / Three Evolutionary Directions for AI E-commerce

Although AI is now widely applied in the e-commerce field, there is still much room for improvement. From my perspective, three evolutionary directions for AI e-commerce deserve attention:

First, multimodal applications. Currently, AI's applications in e-commerce are mostly concentrated in text and images, with relatively few applications in video.

Nowadays, it is common for marketers to use digital human functions with other video templates and product description text to quickly generate product explainer videos or directly input product description URLs to have AI video tools generate brand promotion videos based on the information on the product page. However, both video stability and diversity are lacking.

Considering the increasing importance of short videos in product marketing, multimodal applications will become an important trend for AI e-commerce in the future.

Second, improving AI model capabilities to reduce content hallucinations and randomness.

Currently, for specific questions in the vertical e-commerce scenario, AI still suffers from hallucinations. If there is a suitable mechanism that allows robots to provide answers with evidence, it can attract more people to interact.

At the same time, using the same prompt word to generate copy or images will yield different results, without options for fine-tuning. This uncertainty increases merchants' usage costs.

Third, AI e-commerce applications are moving towards deeper and more segmented needs. For example, more and more merchants are investing in short video advertising, but most can only see some structured data from influencers, such as view counts, follower counts, follower profiles, etc. In the future, AI may truly bring evaluation dimensions to influencer content creation, measuring the match between influencer past video content and merchant brand tones to improve short video advertising efficiency.

For example, there is also much room for AI applications targeting specific needs of consumers in different categories. Domestic efforts in this regard have already begun, such as Turing Authentication, which uses AI authentication for various sneakers and beauty products.

Although there are still many shortcomings in AI's current implementation in the e-commerce field, it must be said that e-commerce is one of the best scenarios for AI's implementation. Its mature digital infrastructure, coupled with its natural proximity to transactions, can not only maximize AI's value but also allow AI companies to move towards commercialization faster.

Perhaps the impact of AI on e-commerce will far exceed our imagination.