On September 4, 2025, at the Alibaba International Station Summit held in Las Vegas, Zhang Kuo, President of Alibaba International Station, demonstrated the AI search engine Accio to over 3,500 attendees. This AI agent can generate development plans based on product concepts, integrating market analysis, supplier auditing, and bulk inquiries, compressing processes that traditionally take weeks into just minutes.

He emphasized that this is not a "Silicon Valley-style moonshot project," but a practical tool to empower global small and medium enterprises (SMEs) in addressing daily challenges.

"AI is no longer a luxury but a necessity for survival, especially in B2B sectors where high-risk decisions require innovation. User behavior is evolving: keyword searches are insufficient to meet demands. B2B urgently needs AI tools supporting long text and multimodal capabilities to navigate global trade. This is why we're comprehensively upgrading this 26-year-old business to help SMEs maintain competitiveness in a world where speed, precision, and adaptability are crucial," Zhang stated.

Zhang Kuo has led Alibaba International Station since 2017. The e-commerce platform currently serves over 50 million SMEs across more than 200 countries. Alibaba International Station is transforming from initial information matching to comprehensive global trade infrastructure covering transactions, payments, and logistics through continuous AI tool iterations.

In this interview, Zhang shared his insights on AI empowering SMEs, reshaping cross-border e-commerce, and changes in foreign trade talent structures and organizational culture in the AI era.

The following is an edited transcript of the conversation:

**Q: Today's presentation highlighted Accio, an AI agent that automates procurement processes. Could you share Accio's current usage scale and how it helps SME sellers reduce procurement barriers and improve efficiency?**

**Zhang Kuo:** Both Alibaba International Station and Accio significantly reduce cross-border procurement difficulties. Cross-border procurement must overcome time zone differences, language barriers, and lengthy, repetitive processes. Previously, using search engines would yield 10,000 products for one keyword. You'd click on appealing items, chat briefly, find they don't work, then search for the next one. Contacted merchants might be out of service area or offline, with requirement communication alone taking a week.

Accio proactively handles many tasks for buyers. Among over 100 application scenarios, two categories are most important: First is "what to sell/source." When you have an idea, traditional methods require market research, product design, and converting designs into executable components—a lengthy cycle.

Second is "who can make this." When you have a prototype idea and need OEM factory partnerships for production, how do you completely match your requirements and ideas with suppliers? This was previously complex because your product might not exist globally. The factory you seek must provide optimal services based on your needs while meeting specific local requirements.

Currently, these basic capabilities are quite mature at the platform tool level. Accio, as an AI-native application, allows us to start with original ideas and use this product to refine and validate go-to-market capabilities. Once these capabilities prove viable and scalable, we integrate them back into Alibaba International Station's main platform.

For example, in September this year, Alibaba International Station's PC version fully launched AI search functionality. This means users can search using natural language, with the system intelligently understanding and matching requirements. For instance, users can input long phrases like "portable power station, suitcase-style, overload protection, under 25kg." Traditional search engines cannot accurately match such queries, but our system intelligently breaks them down into multiple elements for precise matching.

After AI search launch, user-side metrics showed significant growth. Conversion effects alone improved by at least 10%.

**Q: The 10%+ conversion growth—does this refer to inquiry conversion or payment conversion?**

**Zhang:** Both conversion aspects are included. In B2B transactions, average order values are high (our platform's average unit price is approximately $3,000), so users rarely place "silent orders" directly. The entire conversion process typically involves two steps: First is opportunity conversion—converting buyer inquiries into genuinely interested, effective communications. This step directly relates to search matching precision, as better matching faster facilitates connections between parties.

The second step is payment conversion—buyers and sellers finalizing payments after communication. This step, beyond matching accuracy, is influenced by multiple factors including product pricing, logistics delivery time, and fulfillment capabilities.

Our entire order volume growth this year mainly benefited from these two optimizations: On one hand, we optimized front-end matching and search through AI, enabling more users to find needed products more simply and precisely, expanding traffic "openings." On the other hand, we made substantial improvements to payment and logistics infrastructure. For example, in Latin America and Middle East regions, we integrated more local mainstream payment methods rather than relying solely on international card organizations like Visa and Mastercard. This significantly improved local buyers' payment convenience and willingness, further enhancing final conversion rates.

**Q: How much new user growth has AI tools brought?**

**Zhang Kuo:** From overall data, our platform's search engine traffic grew over 20% this year. Search growth is the most front-end, core metric in supply-demand matching. This growth partly comes from more new users starting to use the platform, and partly from significantly improved user retention—previously, users might not return after unsuccessful searches, but now they can quickly find needed products and are more willing to return for repeated searches and purchases.

Specifically for Accio, as a newly launched AI-native application, August monthly active users reached 1.5 million, expected to exceed 2 million in September, with current monthly growth rates maintaining 50%, still in early growth stages. Particularly noteworthy is that Accio users overlap only 30% with traditional Alibaba International Station users, presenting considerable incremental space. We plan to further integrate Accio's agent capabilities into Alibaba International Station in October and November this year.

**Q: AI hallucination is a common industry challenge. In practical application scenarios like merchant search, hallucinations could cause economic losses. How do you ensure data accuracy and reasoning rationality?**

**Zhang Kuo:** In specific search scenarios, demand-side hallucination problems may not be the primary contradiction. Actually, understanding user needs through natural language and intelligent decomposition, then matching merchants, is more accurate than traditional keyword searches.

Traditional search modes have inherent "hallucination" problems. Previously, merchants would repeatedly post products and modify titles to improve keyword rankings, even adding contradictory descriptions (like simultaneously marking "black" and "white")—this could be called "artificially created hallucinations." Now, with large model-based understanding, user queries are longer and more precise, allowing systems to better grasp real needs.

We mainly improve accuracy through two approaches: First, we guide merchants to integrate all product information—including images, text descriptions, store and factory materials, product catalogs, and even new product plans—through multimodal methods into dedicated knowledge bases. AI matches buyer needs based on these complete knowledge bases, naturally achieving higher accuracy. Second, thanks to AI's significant progress in reasoning and multimodal capabilities this year, we can now decompose complex tasks into multiple sub-tasks, process them with different models, then use a main task to coordinate processes and verify results. With long context and better models, error probability has greatly decreased.

We adopt different processing methods for different problem types: For questions with clear answers, such as mathematical calculations or code execution, verifiable standard answers exist. For creative questions, like designing clothing for ADHD children, while no unique answers exist, we can evaluate output quality from multiple dimensions including existing market products, design uniqueness, and user considerations, then continuously optimize models through actual conversion data. Most critically, for the third category—when buyers and sellers must finalize transactions—we adhere to the principle of "don't participate if uncertain." Once requirements exceed system knowledge scope, we immediately transfer to human handling. After human responses, new knowledge updates FAQ systems, enabling continuous machine learning. Through this human-machine collaboration approach, we can strictly avoid transaction risks that hallucinations might bring.

**Q: With AI participation, how much has e-commerce penetration in international trade improved? Do you believe AI's efficiency improvements will have disruptive impacts on global trade infrastructure? What stage is current development at?**

**Zhang Kuo:** Temporally, AI applications in cross-border trade have just begun. But I strongly agree with your view that AI indeed has disruptive potential. People often say one AGI (Artificial General Intelligence) standard is whether it can affect 10% of global GDP. Global trade scale is approximately $30 trillion. While portions like bulk trading may not necessarily be completed by SMEs, our focus is precisely SME trade. This year, our platform's order volume grew 33%, largely because more people are participating in cross-border trade.

Previously, enterprises might only find local agents or large wholesalers for procurement. Now, leveraging global supply chains and AI to lower barriers, international trade participation thresholds have greatly decreased. Today, everyone can connect to globally richer supply resources through AI—this is undoubtedly a future trend. As participation increases, penetration rates will naturally continue rising.

**Q: Looking ahead 3-5 years, which cross-border trade pain points do you think AI can best address?**

**Zhang Kuo:** Cross-border trade's essence is supply-demand matching and transaction fulfillment. Currently, AI mainly plays roles in supply-demand matching: helping decide what products to make, what goods to find, and locating suitable partners. After matching completion, buyers and sellers need communication, where AI can also play significant roles—managing multiple suppliers from buyer perspectives, serving numerous buyers from seller perspectives. Our AI-assisted communication system already achieves 80-point levels, surpassing 80% of human customer service, though gaps remain with the top 20%, but improvement space is vast.

In transaction fulfillment, AI can provide more suitable payment solutions and financial products based on buyer-seller profiles. Regarding logistics, while point-to-point delivery efficiency is already high, cross-border logistics still contains many complex segments requiring human participation, such as consolidated shipping (integrating multiple packages into one shipment to the US). These segments requiring freight forwarder and shipping company coordination offer AI significant optimization space.

Ultimately, we hope to achieve this: users only need to voice requirements to phones, confirm, then systems automatically distribute needs globally, completing product manufacturing and shipping. While order volumes may need gradual growth processes, we believe we'll soon see: cross-border procurement processes originally requiring 27 steps becoming much more efficient through AI assistance. Unlike B2C e-commerce where AI on platforms like Amazon is merely icing on cake, in B2B sectors, its impact will be revolutionary.

**Q: Today's Accio demonstration mainly focused on front-end matching functions. You mentioned back-end logistics consolidation applications. In payment segments, what specific optimization scenarios does AI offer?**

**Zhang Kuo:** Regarding payments, we're mainly doing two things: First is connecting global payment networks, enabling buyers everywhere to pay conveniently using local mainstream payment methods while allowing Chinese suppliers to smoothly receive RMB. Currently, this aspect has achieved good progress—most user experiences are as simple as online shopping.

Second is credit system construction. As business develops, more buyers need credit terms. This requires risk assessment and credit limit management through AI based on buyer historical behavior data. Our new product collaborating with Silicon Valley startup Slope will soon launch, utilizing AI capabilities for risk control assessment—analyzing buyer past transaction behaviors and other information to precisely evaluate default risks. AI still has significant development space in this area.

**Q: Another concern is compliance. Many Chinese companies going overseas find addressing various regional compliance requirements very complex. Is this mainly solved through AI or still requiring human involvement?**

**Zhang:** We provide sellers with 4 intelligent assistants (Agents). Today's demonstrated Accio mainly targets buyers, while seller-designed assistants we call "Business Assistants," segmented according to work scenarios.

Seller time-consuming scenarios mainly include: product listing, advertising placement, customer communication and existing customer maintenance. The fourth, which we've heavily invested in from the beginning, is the risk control Agent, directly related to compliance issues.

Specifically, risk control Agents mainly help sellers solve several practical problems: First is infringement risks. Traditional models meant sellers might only know about product infringement upon receiving lawyer letters, forcing passive delisting. Now through AI, we index global data, conducting multi-dimensional scans from day one of product listing. Even if we can't 100% avoid risks, we can significantly reduce infringement probability.

Second is qualification certification. Different market requirements vary greatly. Risk control Agents inform sellers before expanding into certain markets about required certification credentials, avoiding inability to sell due to missing qualifications. This also includes transaction dispute handling. For instance, when Chargebacks occur, Agents can assist sellers in drafting appeal letters and preparing appeal materials.

These are characteristic difficulties distinguishing cross-border trade from domestic trade, but fortunately, related rules and knowledge are quite clear, making them very suitable for AI efficiency improvements. Actual effects are obvious: platform deductions against sellers for compliance issues have dramatically decreased. As long as sellers follow Agent guidance, they can basically avoid such penalties. This significantly reduces operational costs and risks for both platforms and sellers.

**Q: Alibaba was once famous for its "Chinese Supplier Iron Army" corporate culture. In the AI era, how has organizational culture changed? How will AI change foreign trade industry talent structures? Which positions and capabilities will be more valued?**

**Zhang Kuo:** From Alibaba International Station's perspective, our AI applications can be viewed across three levels.

First is AI Native applications, like Accio—from its first day, all interaction paradigms and technical architectures were completely designed based on AI.

Second is AI Plus existing business enhancement. The platform has mature ecosystems and business models, so we more prudently advance AI scaled applications. Usually, we conduct cutting-edge trials in innovation units, then integrate into main sites after successful validation. Simultaneously, we help platform merchants upgrade business methods, making their operations simpler.

Third is AI Inside organizational empowerment. We focus on how AI improves everyone's efficiency within enterprises. Now, every position—whether industry operations, technical development, sales, user growth, marketing, customer service, or risk control—has corresponding AI empowerment targets and metrics. AI doesn't just serve technical or sales teams but comprehensively integrates into all functions.

For example, sales colleagues previously spent considerable time organizing customer materials, making outbound calls, and recording follow-up processes. Now AI can automatically complete much repetitive work, enabling newcomers to quickly reach experienced efficiency levels, allowing sales to focus more on serving customers and building meaningful relationships. For technical teams, previously each quarter could implement approximately 100 product iterations, but accumulated demands might be 300. Now, leveraging AI programming assistants and other tools, R&D throughput significantly improves. Ideally, the number of ideas completable in the same time can double, dramatically improving product iteration efficiency.

AI is changing every position's working methods. It might be your copilot or autopilot, but goals are always liberating people from repetitive labor to do more valuable work. Therefore, for young people, I suggest focusing more on capabilities requiring human judgment, creativity, and emotional connections—these are currently difficult for AI to replace. Simultaneously, actively learn how to collaborate with AI, utilizing tools to improve your decision-making and problem-solving efficiency.

**Q: For resource-limited companies wanting to go overseas, especially regarding how to use AI tools and Alibaba International Station to quickly test international markets and find opportunities, what suggestions do you have?**

**Zhang Kuo:** Based on customers, roughly four categories exist, each with different suggestions. For individual operators or small teams, AI can completely enable quick startup. For example, a college student entrepreneurship team led by Liu Shiqi with only 6 people, through AI for product creative design, automatic inquiry responses, and intelligent placement, focusing on slipper products, now achieves annual export volumes in the tens of millions. They leverage both China's supply chain advantages and AI to amplify creative capabilities, achieving business scaling.

For factory-type sellers, the core is "digitalizing" your manufacturing capabilities. Previously, production lines, processes, and certificate information were isolated. Now, uploading product catalogs and certification information through platforms enables global buyers to precisely discover your advantages. This is actually similar to "RAG" (Retrieval-Augmented Generation) concepts in AI—using your factory's complete materials to construct more retrievable and understandable "factory profiles."

For traders, advantages lie in product selection and supply chain integration. AI can help you quickly generate new ideas and discover new opportunities, enabling more efficient category expansion.

For sellers already doing domestic e-commerce, overseas barriers have significantly decreased. You can completely convert Taobao and Tmall product information into multilingual versions with one click. Platforms help solve language and time zone issues, even previously complex matters like receiving foreign exchange and cross-border logistics—infrastructure is now complete. Essentially, you only need one decision: continue focusing solely on domestic markets or go global? From imagination space and profit level perspectives, global markets undoubtedly offer greater opportunities.

**Q: After discussing many buyer paradigm changes overseas, from seller perspectives, what cross-border trade paradigm changes is Alibaba International Station bringing through AI?**

**Zhang Kuo:** To better help merchants understand and utilize platform capabilities, we recently summarized complex overseas processes into new paradigm iterations of "Four Venues."

First is the Shelf Venue. Traditional methods relying on keyword stuffing and repeated product posting are outdated. Now AI through natural language and multimodal matching can precisely understand buyer needs, so merchants need to return to fundamentals—completely uploading real factory capabilities, product data, and enterprise information, letting AI help with most precise matching.

Second is the Business Venue, B2B's characteristic domain. Every inquiry is crucial because 70% of B2B merchant revenue comes from existing customer repurchases. AI here not only assists customer communication but helps manage customer relationships, diagnose sales processes, and even improve entire business team efficiency.

Third is the Marketing Venue. Previously relying on "traders" manually selecting keywords for bidding had limited efficiency because buyer search queries are extremely diverse. Now AI should decide what products to promote, what keywords to use, and whom to target—machine generalization capabilities exceed human capabilities by ten thousand times. People only need to set target ROI and let AI execute.

Finally is the Fulfillment Venue. Global supply chains are now more diversified: US local warehouses can achieve 2-3 day delivery; domestic spot goods or light customization products can deliver within 14 days; heavy customization via ocean freight costs less but requires longer cycles. Merchants can choose most suitable fulfillment methods based on product characteristics and buyer needs.

These "Four Venues" constitute a complete closed loop, enabling merchants to receive empowerment at every stage from product display, opportunity conversion, marketing promotion, to final fulfillment.