AI e-commerce is "converting exports to domestic sales"

In the complex environment of cross-border e-commerce, AI has begun to help merchants solve multilingual and multicultural problems, and play a role in advertising, intelligent customer service, etc. This article will explore the current status of AI applications in cross-border e-commerce, and how it can provide value to merchants, which may eventually feed back to the development of domestic e-commerce.

"The advanced domestic e-commerce model is simply a dimensionality reduction attack when it reaches overseas." This sentence has become a public consensus.

In the past few years, Temu and Tiktok have copied the domestic e-commerce model overseas and achieved overwhelming success, while AliExpress and Shein went overseas even earlier. However, with the advent of the big model era, this process seems to be reversed: AI, which has not been widely used in domestic e-commerce, is taking the lead in blooming and bearing fruit in cross-border e-commerce, and eventually feeding back to domestic e-commerce.

1. The large model is slow to be implemented in domestic e-commerce

"The era of AI e-commerce has just begun. It is an opportunity and a challenge for everyone." At the end of November last year, Jack Ma's speech on Alibaba's intranet excited countless people, and everyone is looking forward to the arrival of a new era of e-commerce. However, to this day, neither merchants nor consumers have felt how much impact the AI ​​big model has on their sales or purchasing experience. The era of AI e-commerce seems to remain only in people's imagination.

Looking back two years ago, when ChatGPT was just becoming popular, everyone was discussing which industry it would first bring about change. In the end, the answer pointed to e-commerce, especially the customer service scenario in e-commerce.

The reason is simple. Customer service is indeed a labor-intensive industry that urgently needs to use new technologies to reduce costs and increase efficiency. At the same time, the customer service scenario is relatively closed, and model training is easier to achieve than in an open scenario.

But judging from the results, all the domestic companies that make customer service robots on the market have not shown significant performance improvements after transforming their products with large models.

A product manager of the intelligent customer service of an e-commerce platform explained the reason to Leifeng.com. After-sales service is a very closed scenario. Users will only come to after-sales service when they encounter problems during the purchase process, which requires the responses provided by intelligent customer service to be very accurate. Even in the generalized field, large models will have "illusion" problems. In the refined scenario of after-sales service, it is even more impossible for the platform to let it face the user to generate content arbitrarily.

The platform it is on has fallen into some traps before. Although very strict risk control measures have been taken internally, the intelligent customer service sometimes still makes some inappropriate promises to users. For example, it promises to compensate users when it is clearly not the merchant or the platform’s fault, and the user finally takes this promise to the platform to redeem it.

In addition, the intelligent customer service was based on the previous generation of artificial intelligence technology. The model capabilities were not as strong as they are today, and inevitably there were many places that required human intervention. After upgrading to a large model, the workflow also needed to be changed accordingly. All these led to the fact that the implementation effect of the large model in the intelligent customer service scenario was not as good as expected.

Pinduoduo also began to use big model-like technology to reduce costs and increase efficiency in the customer service process around 2020. However, it adopted a violent aesthetic approach to solving the problem. Once it detected user dissatisfaction, the platform intervened directly to only refund the customer service robot, without caring whether the big model made the conversation between the customer service robot and the user smoother. This has achieved good results in business, but it is hard to say that it is a technical victory.

The aforementioned product manager believes that large models may be more suitable for open domain conversations, such as chatbots. After all, in these scenarios, users do not have such high expectations for it.

For example, a number of AI technologies are being put to good use at the Paris Olympics. It is reported that the first large-scale model application in the Olympic field was developed by Alibaba's international AI team and will be open to official commentators of the International Olympic Committee to assist in the commentary of various events and provide more professional and interesting support for the Olympics.

In addition to after-sales service, AI shopping guide products based on large models have also not developed as expected. An industry insider said, "Whether AI can do a good job in pre-sales service depends on whether the platform's content ecosystem can be established, rather than just relying on itself to change."

In addition to after-sales and pre-sales scenarios, advertising is also an application scenario of AI big models that major e-commerce platforms are currently trying hard to implement.

In the earnings call in May this year, Alibaba disclosed the latest progress of the site-wide promotion function: the site-wide promotion is in the small-scale customer testing stage... Taobao is currently adjusting the algorithm model, taking a longer time to train the model, and based on more customer data, to improve the efficiency of guaranteeing customer ROI.

Internally, it is determined that it will take time to add customers and match traffic among different industries and user groups, and it is stated that "it will take another 12 months to see significant revenue growth from the full-site promotion."

In other words, AI can only solve the needs of "from scratch" in the e-commerce field at this stage, and it will take time to achieve "from existing to excellent". Therefore, it is not surprising that AI can first bear fruit in cross-border e-commerce.

2. Be the first to bear fruit in cross-border e-commerce

Compared with domestic e-commerce, cross-border e-commerce has longer links, more links, and faces more diverse challenges.

The first is the issue of multilingualism and multiculturalism. Different countries have different languages ​​and time zones, which brings huge challenges to merchants in writing product descriptions and arranging customer service working hours.

At the same time, overseas marketing is more difficult, traffic acquisition costs are higher, regulatory compliance requirements are very complex, and there is a severe shortage of relevant professional talents. This gives AI more room for use, and businesses get a stronger sense of value from AI.

For example, merchants often encounter credit card chargebacks when doing cross-border business. That is, after a user places an order and pays with a credit card, if they have any doubts about the transaction, they can call the card issuer to file a chargeback.

At this time, the card issuer will send an email to the merchant to inform the merchant of the user's concerns. If the merchant replies to the email in a timely manner and provides relevant evidence to defend, the credit card institution will still pay the merchant in the end.

But the reality is that many businesses don’t know how to reply to emails because it involves very complex financial compliance knowledge.

Today, Alibaba International can use AI to help merchants reply to this email, and the approval rate of credit card companies is very high, even higher than that of manually written emails. According to official disclosures, this function can save the platform about 20 million yuan per year.

Another interesting example is digital human live streaming. For a while, domestic e-commerce platforms had a strict ban on digital human live streaming. Once a merchant was found to be using digital human live streaming, the account might be directly blocked. However, digital human live streaming is much more accepted overseas.

The reason behind this is that the essence of live broadcasting is to build trust between users and anchors to improve conversion efficiency. Domestic users first came into contact with live broadcasting from real anchors, and switching to digital human live broadcasting is a regression in user experience for them, which is naturally difficult to accept. Moreover, after several years of development, the supply of domestic live broadcast professionals is relatively abundant, and there is not much demand for digital humans.

Therefore, when JD.com launched the digital man "Dong Ge" live broadcast to sell goods, it was criticized by many users. "At that time, this live broadcast was preheated for a long time online, which whetted everyone's appetite. Unexpectedly, it was a fake person who came in the end. If it wasn't for Lao Liu chatting there, who would have nothing to do to come to JD.com to watch cartoons?"

In contrast, overseas, many users have not yet been exposed to live broadcasting, and they are more receptive to digital people. Moreover, overseas live broadcasting professionals are also very scarce. When TikTok first started live broadcasting e-commerce in Southeast Asia, a TSP agency poached all the employees of a massage parlor in order to build a live broadcast team.

The agency explained, "A good anchor needs to have a good appearance, a generous personality and eloquent verbal expression skills, and the service staff in massage parlors best meet these three criteria."

In addition to these, AI big models also have broad application prospects in the search, advertising and promotion links at the lower level of cross-border e-commerce platforms. At a recent media communication meeting, Zhang Kaifu, vice president of Alibaba International, said that the progress of using big models to transform search, advertising and promotion in China is very slow. Big models can only play a role of icing on the cake, and cannot subvert the underlying architecture. On the contrary, the gains from using big models to transform search, advertising and promotion abroad are much greater.

Because traditional search and promotion engines are not built based on multimodal data, but on behavioral data, domestic e-commerce user behavior data is already rich enough, and adding a little multimodal information is not very helpful. However, user behavior data in many overseas countries is still relatively sparse, so adding multimodal data in this case is very helpful in improving user conversion rates.

However, Zhang Kaifu also mentioned that improving the accuracy of e-commerce platform search results through large models involves two issues - understanding user intent and extracting product features.

There are two challenges here. The first is how to balance the accuracy of search results and the response speed. "If you use some larger models, the matching degree may be improved, but the response speed may not be guaranteed. At this time, you have to make some trade-offs." The second is how to better abstract and express product features. This may require the reconstruction of the entire product release process, rather than just superimposing a set of models on the existing basis.

This is also a topic that Alibaba International needs to think about and explore further.

3. The biggest challenge is to get merchants to use it

The multi-language, long links, and many stages of cross-border e-commerce have given AI more room to play. However, as an advanced productivity tool, AI itself has a certain threshold for use. How to get merchants to use it is also a question that must be answered in the process of its popularization.

Looking at the many advanced productivity tools in the history of business development, there is often a paradox: small and medium-sized businesses have a greater demand for advanced productivity tools, but due to their own lack of funds, talents and other aspects, their use often lags behind large enterprises. As a result, large enterprises think they are useless, and small enterprises cannot use them. How can we get out of this vicious circle?

Zhang Kaifu believes that the AI ​​technology revolution is different from previous technology revolutions. AI can interact with natural language, and its use threshold is naturally lower than previous IT tools. When facing AI tools, large enterprises need to change their internal organizational processes to adapt, which is more difficult to use. On the contrary, small companies will be more flexible.

In terms of product design, Alibaba International also seeks to integrate AI into business processes so that merchants can use it without any feeling. Taking the credit card chargeback defense email introduced earlier as an example, when the merchant downloads the document in the background, all relevant information will be automatically filled in the document, and the merchant does not need to perform any unnecessary operations.

In addition, compared with domestic e-commerce, cross-border e-commerce has a higher proportion of hosting business. Under the hosting model, merchants, regardless of size, are provided with operating services by the platform. After the platform obtains authorization from the merchant, it can apply AI capabilities to all aspects of its business. This makes the use rate of AI tools in cross-border merchants much higher than that of domestic merchants.

According to official disclosures from Alibaba International, its generative AI has empowered more than 500,000 merchants in more than 40 e-commerce operation scenarios. Since its launch in November last year, the daily call volume of Alibaba International's generative AI has doubled every two months, and now the average daily call volume has exceeded 50 million times, and the descriptions of more than 100 million items have been optimized through AI.

Alibaba International has conducted A/B tests on a considerable number of the more than 40 business scenarios where AI has been implemented, and found that AI not only helps merchants save labor costs, but also improves the click-through rate and conversion rate of products by improving the quality of content. In different business scenarios, the efficiency improvement brought by AI varies. For example, in the pre-sales shopping guide and reception links, the conversion rate can be increased by up to 30%, but in some scenarios, it is only increased by 1-2 points.

As we all know, AI is a beast that is fed with data. Only when it is applied to business scenarios and continuously absorbs various data for training can it grow faster. However, due to the poor performance of AI in the early days and the fact that domestic e-commerce is mature enough, e-commerce merchants do not have a strong sense of its value. Therefore, AI is not widely used in e-commerce business, which in turn restricts its growth.

Today, the cooperation between cross-border e-commerce and AI has successfully broken this deadlock. I believe that in the future, after AI tools are polished and mature in cross-border e-commerce, they will also be able to better feed back to domestic e-commerce. The history of domestic advanced e-commerce models being copied to cross-border e-commerce may really be reversed.

Author | Liu Wei Editor | Lin Juemin This article is written by the author of Operation Party [Leifeng.com], WeChat public account: [Leifeng.com], original/authorized to be published on Operation Party, and any reproduction without permission is prohibited.

The title image is from Unsplash, based on the CC0 protocol.