This episode of Alicast comes at a pivotal moment when artificial intelligence is rapidly becoming integral to global business operations. As companies across industries explore how AI can enhance efficiency and drive growth, Alibaba International Digital Commerce Group (AIDC) has taken the lead in applying these technologies to empower small and medium-sized enterprises (SMEs) in cross-border e-commerce.

Host Selina Zhang takes us inside Alibaba’s headquarters in Hangzhou to explore how artificial intelligence is not just enhancing business efficiency but fundamentally reshaping global commerce.

Joining the conversation is Kaifu Zhang, Vice President and Head of AI Initiatives at AIDC. Through the lens of AI-driven solutions—ranging from advanced translation engines to multimodal customer service tools—Kaifu offers a powerful narrative on how technology can lower barriers, enabling even the smallest enterprises to thrive on a global scale.

Tune in to discover how AI will shape business operations over the next decade—and, most importantly, why the smallest enterprises stand to gain the most, and how companies that embrace these AI-driven tools today will be best positioned to lead the future.

Below is a transcript of this Alicast, edited for clarity and brevity

Selina Zhang:

Welcome to Alicast, where technology and business converge. I’m your host Selina Zhang. Over the last year or so, we have been hearing about AI non-stop. In this episode from Alibaba’s headquarters in Hangzhou, we are diving deep into how AI is revolutionizing global business with our special guest, Kaifu Zhang, Vice President and Head of AI Initiatives at Alibaba International Digital Commerce Group. Kaifu. Thank you for joining us on Alicast.

Kaifu Zhang:

Selina, it’s great to be here.

Selina Zhang:

So, we are going to talk about the buzzword AI. It feels like we are living through this historical shift where AI is suddenly everywhere. Can you share how your background has influenced your approach to AI and what personal experiences or insights have shaped your vision for its future?

Kaifu Zhang:

Thank you, Selina. My background was initially in computer science. Later, I started my academic career at the intersection of AI and economics as a professor, firstly in China and then in the US at Carnegie Mellon University. Then, I left academia and joined Alibaba as an e-commerce operation guy. So, I was in charge of Taobao’s operation and, later, our cross-border commerce operation for many years. After AI came about, I picked up this sort of interest in the subject long ago. I was kind of connecting the dots by finding a new initiative at the intersection of AI technology and also its e-commerce application. And I’ve had a really great time in the past one and a half years applying AI to e-commerce.

Selina Zhang:

For our listeners who might not be familiar with AIDC, Alibaba’s international platforms like Aliexpress and Alibaba.com, meanwhile, leading the chart in AI innovation in their own way, like Kaifu just mentioned. So, could you share more about AIDC’s recent strides in this field, like what your team is doing daily?

Kaifu Zhang:

Selina, as you mentioned, AIDC is the collection of all the Alibaba and e-commerce businesses operating out of China. Actually, we have multiple platforms. As you mentioned, we have Aliexpress, a cross-border platform, Lazada in Southeast Asia, Trendyol in Turkey, Daraz in South Asia, also Taobao and Tmall Global, etc.

Overall, all these platforms may have different business models and operate in different geographic locations, but they share a common goal: leveraging technology to enable small and medium-sized firms to sell globally. Now, many of them are small, with just three to six people,  and often don’t have good access to talent. While the Internet gives them the ability to reach global markets, they still face tremendous challenges in understanding various foreign markets, gaining consumer insights, and overcoming language barriers. Even tasks like preparing product listings can be tough. That’s where AI plays a key role. We found it to be extremely instrumental in lowering the barrier and making it easier for small enterprises to compete and grow globally.

Selina Zhang:

What emerging AI technology are you most excited about for the merchant-facing AI you just mentioned? And how do you see them shaping Alibaba’s global operations?

Kaifu Zhang:

When engaging in a global e-commerce operation, a lot of the merchants have trouble preparing product listings. Basically, how do you describe your product to consumers from a different country? Sometimes from a different culture? They speak different languages and have different preferences for product descriptions. We also need to localize the content according to the merchants’ market. Merchants also have to face marketing, customer service, and compliance challenges.

In all these scenarios, we are developing AI tools to help merchants better serve consumers, right? For example, AI could generate or adapt the product listing they posted online to fit local consumer preferences and linguistic habits better. And the marketing messages they develop can also be generated by AI to increase conversion rates. Customer service can be helped by adopting an AI chatbot or sometimes a customer service Copilot. AI can also automate the compliance process.

If you look at all these different applications, it’s essentially a tool provided for the merchants to help them better navigate the complexities of global commerce. But at the end of the day, whatever content the AI model generates is presented to the consumers, helping them to make a shopping decision in a more authentic linguistic experience and to improve the quality of decisions they ultimately make.

Selina Zhang:

Right now, a lot of the way people talk about AI is something like a hammer looking for a nail. Here are large language models; how can we use them? So you just mentioned that most of what you have done is for the SMEs. So, I’m interested in whether the benefits of AI are equally distributed across industries or business sizes. Do specific sectors like SMEs stand to gain more?

Kaifu Zhang:

Yeah, I think you’re exactly right about the metaphor of the hammer looking for a nail. If you look at the AI industry now, I think in our scenario, there are really a lot of nails looking for a hammer. So the language barrier I mentioned earlier, the lack of talent, and a lot of the SMEs facing the complexities of compliance are all nails looking for a hammer. So, if you look at the typical operation of, for example, a small enterprise before the age of AI, it can be really challenging, correct? The product listing they prepare contains a lot of grammatical mistakes, right? So, they could have designs with foreign characters and present those marketing creatives to their consumers without modification.

Customer service suffers from both a language barrier and time difference. So when the consumers in the US ask about a product sold by a seller in Vietnam, the guy might not even be sitting in front of the computer because of the time difference. All of these are true nails.

From a consumer experience perspective, the current model isn’t ideal. While the internet allows even small sellers to reach a global audience, it doesn’t always deliver the best experience for consumers. So basically, our approach focuses on addressing a very specific problem in marketing, such as product design, customer service, and compliance, by asking, How can AI help? This often involves creating content generation processes or building digital agents to automate tasks for small merchants. It’s really in nails looking for a hammer. Interestingly, it turns out that the smaller and less equipped the merchants, the more they stand to benefit from AI solutions.

Those are the sellers who are going to market with grammatical mistakes, who are going to market with no customer service agent handling requests from the consumers. And with AI, they really went from 0 to 1. It is really different between having and not having, right? So those are actually the merchants who benefit the most and stand to gain the most from the deprivation of AI technology.

Selina Zhang:

I’m curious about like why AIDC and Alibaba are so well positioned to be leaders in merchant-facing AI, why Alibaba could solve this kind of problem well, and how you ensure that the AI tools on your platform can be relevant, innovative and ahead of market trends？

Kaifu Zhang:

Two key factors in addressing this question are domain knowledge and the data feedback loop.

First, domain knowledge is essential. In many e-commerce applications, you can’t take the large language model out of the box; they must incorporate specific expertise. With years of accumulated data, model training, and fine-tuning become more targeted, addressing specific e-commerce challenges effectively. That is a broad statement.

Second, the data feedback loop plays a crucial role. For example, when we deploy chatbots or translation tools, we analyze tens of thousands of chat sessions daily. Feedback from these sessions—whether customers are satisfied, abandon their carts or complete purchases—helps us refine the model and enhance product design. This iterative process ensures the tools become more effective, benefiting both merchants and consumers. So, those are the two main factors that allow us to further iterate the tools and make them more efficient and effective for our platform merchants.

Selina Zhang:

What are the five or the three most used AI features on AIDC platforms?

Kaifu Zhang:

So, I’ll give you some examples, starting from the simplest to some of the more complicated AI features.

One of the most essential features in global e-commerce is translation. When sellers list products in their local language, the content often needs to be translated into 10 to 20 languages. Relying on human proofreading at this scale isn’t feasible, so AI-powered translation is essential.

Our new translation engine, built on a large language model, goes beyond word-for-word translation. It rewrites product information, capturing the nuances of each language. As a result, the content quality on the platform has improved, particularly in less common languages, leading to a 5% increase in product listing conversion rates. With clearer product descriptions, consumers are more likely to make informed purchase decisions instead of abandoning their carts.

Another example is our AI-driven chatbot, which enhances customer service before and after purchases. Pre-sale inquiries—such as questions about product features or compatibility—are well-suited for large language models using retrieval-augmented generation. The chatbot accesses the product database to provide accurate answers based on the information merchants supply. This pre-sale support boosts merchant conversion rates by up to 30%, highlighting the importance of real-time responses in driving purchase decisions.

That’s another thing I would call low-hanging fruit that we developed in the first months by applying AI for our cross-border sellers. The third application, which is more recent, uses the so-called multimode models. This is a scenario that we use when dealing with customer returns.

Imagine ordering a dress online, but it arrives in the wrong color. So, what do you do? Naturally, you’d want a refund, but platform rules often require returning the product. This process is costly and inconvenient, as you must ship the item internationally, incurring high fees.

For consumers, it means the hassle of visiting the post office and waiting for the refund, while merchants face even greater challenges. After the product travels thousands of miles, it often can’t be restocked—especially for time-sensitive items like clothing—and might not even arrive back in usable condition. Ultimately, the merchant risks losing both the product and the sale, with the only winner being the logistics partner collecting shipping fees.

From the merchant’s perspective, negotiating with the consumer to keep the product in exchange for a partial refund can be more favorable than processing a full refund. That’s where our AI-powered negotiation agent comes in.

The agent initiates by asking the consumer to send a photo to verify the issue—thanks to its multimodal capabilities, it can analyze both the message and the image. For example, if the color is different but the dress is otherwise intact, the agent suggests a compromise: the consumer keeps the dress and receives a 30% refund.

This approach benefits everyone. The merchant retains part of the revenue, avoiding the costs and risks of international returns. The consumer avoids the hassle of shipping and gets partial compensation while keeping a wearable product. In many cases, consumers accept the offer, making it a win-win solution for both parties.

Selina Zhang:

Quite fascinating. So, Kaifu, you mentioned a lot of the tools that can be used before AI, right? So, how do you envision AI continuing to import SMEs on AIDC platforms? Do you foresee validation of all the AI features? Or could it be like a proliferation of special tools catering to specific needs?

Kaifu Zhang:

We anticipate consolidation from both technological and business model perspectives. Over the past year, AIDC has adopted an innovative consignment model designed to lower barriers for merchants. The idea is simple: merchants send their products to our warehouse and sign an agreement with the platform, which handles everything from there.

This model aligns perfectly with our AI tools. Rather than using AI SaaS, where merchants must manage multiple tools and complete dozens of tasks to run an online shop, they only need to ship their products and authorize automation. The platform then manages operations seamlessly, with AI applications running in the background.

This approach streamlines the process, ensuring merchants don’t have to engage directly with AI tools—they simply give permission, and the platform handles the rest. This is a key example of the consolidation we’re seeing in action.

Language models, especially large ones, have advanced significantly in recent months, becoming more powerful and versatile. In our online applications, we adopt a dual approach:

For high-frequency, specialized tasks, we use smaller, distilled models derived from larger ones to reduce costs. However, we rely on a single large model for long-tail applications, fine-tuning it with prompts to handle various tasks efficiently.

You can think of the large model as a versatile generalist, capable of performing multiple tasks across domains. This reflects consolidation in foundational models. Meanwhile, smaller models offer a cost-effective solution for specific, high-frequency needs. The structure of this approach has become increasingly clear, driven by technological advancements that make these models more capable and adaptable.

Selina Zhang:

Yeah, so it could be like you already have 30 or 40 AI tools in your box. Like if I was a merchant, I didn’t know what to happen at the back. So what you have done could be like you can make either large model or smaller models just to deal with all these tasks.

Kaifu Zhang:

We aim to simplify AI tool adoption for merchants, offering around 30 to 40 application scenarios. Some tools, like chatbots, require merchants to decide whether to use them actively. Merchants may opt out if they prefer their customer service team or use the chatbot strategically—such as relying on human staff during the day and activating the chatbot at night.

The process is automatic in other cases, like product description generation or translation. Merchants only need to grant permission, and we handle the rest, improving product listings on their behalf. These automated tools have higher adoption rates since merchants only need to opt in.

We expect further consolidation as we continue integrating AI with our consignment model. This will streamline the tool matrix, making it simpler for merchants by reducing the number of decisions they need to make.

Selina Zhang:

Okay, permission is essential and fundamental for everything.

Kaifu Zhang:

That’s for sure. Yeah, so we always get permission from the merchants, but they don’t have to go through all the hassle of making all the clicks.

Selina Zhang:

Zooming out a bit, how do we see AI evolving in the next 5 to 10 years, particularly in business operations?

Kaifu Zhang:

AI has made remarkable strides in the past three to six months, with technology advancing at an impressive pace. Looking ahead, the next 5 to 10 years present opportunities that were unimaginable before the advent of AI.

One exciting area is AI-assisted content generation. While we currently use AI to create marketing materials and product listings, new developments are emerging. Startups are now leveraging AI to design physical products, significantly enhancing product diversity. Personalized products, once limited to simple customizations like adding initials to a t-shirt, are evolving. With AI pipelines, consumers can now express their unique preferences—uploading photos, selecting art styles, and receiving personalized designs tailored to them.

AI also has the potential to increase product variety exponentially. With advanced translation engines and AI-empowered small businesses, even niche enterprises can compete globally. This will benefit consumers by offering a broader selection of products, including unique and exotic items from different regions, which were previously inaccessible online.

These trends point to a future where AI drives both physical and virtual product abundance, creating new opportunities for merchants and a richer shopping experience for consumers.

Selina Zhang:

It is interesting that everyone can be creative, everyone can be an entrepreneur, and everyone can be a businessman. A lot of exciting things could happen. Yeah, this is so fascinating and wonderful. Thank you so much. It’s a good place to end this conversation. We will see how AI can take us.

Kaifu Zhang:

Thank you, Selina.

Selina Zhang:

Thank you! That’s the show. Thank you for listening to Alicast. Also, there are ways to stay in touch with us, such as by emailing us at editors@alibaba-inc.com. We love hearing from you.

---

Listen and follow Alicast

Apple Podcasts | Spotify

---

“Alicast” is a production of Alizila, the corporate newsroom of Alibaba. It’s produced by Selina Zhang. If you’re already using a podcast app, please follow “Alicast.” If you’re listening on Alizila’s website and want each new episode of “Alicast” delivered to you as they’re published, download any podcast app, then search for “Alicast by Alizila” to follow the show. Thank you for listening, and goodbye for now.