SMBs are overwhelmed by multi-step operational work that drains time and headcount. Global trade processes remain slow, fragmented, and opaque.
For all the focus on large enterprises in AI adoption, World Trade Organization working group documentation notes that SMBs make up nearly 90–95% of all global businesses and account for 60% of worldwide employment. Other NGO findings show they continue to face outsized barriers in trade and finance, which directly align with new heights of AI adoption in those sectors.
The World Economic Forum reports that small businesses particularly struggle with problems that also amplify a “polycrisis” of supply chain-shaking economic shocks and other concurrent global events: limited access to capital, complex regulatory requirements, inadequate digital infrastructure, and the difficulty of navigating international trade rules as smaller entities.
Numerous reports highlight the severe financial challenges faced by SMBs. The 2025 Federal Reserve Small Business Credit Survey notes 75% cite rising costs of goods, services, and wages as the top issue, with 56% struggling to pay operating expenses and 51% facing uneven cash flows. Financing needs drove 59% to seek new credit, primarily for operations (56%), yet 24% received none, and 36% received only partial approval.
But financial strain is only one side of the challenge. Even as SMBs grapple with rising costs and restricted access to capital, they also struggle to adopt the very technologies that could relieve these burdens, particularly AI.
Per SMB Group and Workday’s 2025 AI Trends report, SMBs face major adoption barriers — 48% say they don’t see clear business relevance, 47% cite security and privacy concerns, 28% point to high costs, and 25% report lacking the skills or resources needed to implement AI.
In a recent episode of AI in Business, Emerj Editorial Director Matthew DeMello sat down with Kuo Zhang, President of Alibaba.com, following his address at this year’s CoCreate 2025 event in Las Vegas, to discuss how SMBs and enterprises alike can adopt agentic AI.
Their conversation highlights two critical insights for SMBs navigating AI that this article will examine in-depth for retail, tech, and SMB leaders:
- Automating end-to-end sourcing and procurement for SMBs: Leveraging agentic AI as a force multiplier, SMBs can automate complex, multi-step operations that once required entire teams.
- Selecting the correct problems as the foundation layer: Applying a three-layer methodology of problem selection, leveraging data and domain knowledge, and model deployment to build effective agentic AI for SMBs.
- Building AI-native applications to enable rapid innovation: AI-native platforms let SMBs redesign workflows, experiment with new models, and accelerate product development without legacy constraints.
Listen to the full episode below:
Guest: Kuo Zhang, President, Alibaba.com
Expertise: SMEs, Artificial Intelligence
Brief Recognition: Kuo has been with the Alibaba Group since 2011, where he previously led the Merchant Services Business Unit — overseeing operations tools, platform development, and ecosystem management. In his current role, his focus remains consistent with the company’s core mission: making it easier for SMBs to do business anywhere. He holds a Master’s degree in High Performance Computing from Tsinghua University, China.
Automating End-to-End Sourcing and Procurement for SMBs
Kuo opens the conversation by explaining that the company’s vision for agentic AI goes far beyond assistants that generate documents or plan travel. Their platform, Accio, launched last year, is now used by more than two million SMBs. It was designed to take over real operational work that usually requires large teams and extended timelines.
He illustrates this with a regional sports event in Latin America that needed to procure hundreds of thousands of items for a multi-country competition. A process that once took four months — identifying suppliers, checking compliance rules, and coordinating inquiries — now takes hours.
Users upload an Excel sheet with items and requirements, and Accio interprets the details, checks import regulations, scans Alibaba’s 200,000–supplier network, generates a vetted supplier list, and even initiates outreach until a final decision is ready.
He also shares a second example involving a solo entrepreneur designing clothing for children with ADHD. What would usually require weeks of market research, design iterations, specification writing, supplier outreach, and prototyping discussions can now be executed by the agentic system.
Kuo describes for the Emerj podcast audience how Accio interprets the initial idea, researches the market, drafts designs and detailed specifications, and contacts suitable manufacturers — all of which previously demanded significant time and coordination.
He draws a stark contrast with traditional search engines, where users type keywords and sort through thousands of results themselves. In contrast, the agentic model lets SMBs upload complete context — designs, specifications, product lists, or full spreadsheets. The system produces an execution plan that users can tweak in plain language, then automatically launches and manages the entire workflow.
Kuo stresses that the experience is fully turnkey: no installation, no setup, and instant usability through a browser. For Alibaba.com, the simplicity he describes is central to the vision: an out-of-the-box agent replaces manual, time-consuming work with an autonomous system capable of running end-to-end business processes.
Selecting the Right Problems as the Foundation Layer
Kuo lays out the philosophy behind how his team builds agentic AI tools, framing it as a three-layer methodology that starts long before any model is trained:
- Choosing the right problem
- Leveraging data, domain knowledge, and operational history
- Reinforcing with underlying models and infrastructure
The first and most essential layer, he says, is choosing the right problem. Many products fail not because the technology is weak, he insists, but because the problem they target is either insignificant or not truly felt by users:
“In our case, the core challenge we are addressing is the complexity of global trade. This is a very real and significant problem — one that represents roughly $30 trillion in global economic activity.
Given its scale, it is a challenge worth tackling. With more than 26 years of experience, we have developed a deep understanding of global buyers and suppliers, and of the pain points they face across markets.”
–Kuo Zhang, President of Alibaba.com
The second layer that Kuo describes is the ability to leverage the data, domain knowledge, and operational history that the company already possesses. With billions of products, hundreds of millions of buyers, hundreds of thousands of suppliers, and a constant stream of daily transactions, the system can learn what a good answer looks like in context.
He stresses that this foundation matters because agentic AI must not only generate an output — that output must be correct, reliable, and operationally sound. That requires careful design, context awareness, and an evaluation framework grounded in real-world activity.
The third layer involves the underlying models and technical infrastructure. Kuo notes that the company uses some of the largest language models in the world and layers its own applications and domain-specific intelligence on top of them. But even here, he emphasizes that the model is not the starting point; it is the final ingredient in a system built on problem selection, domain expertise, and rich data.
Across these layers, Kuo returns to one central idea: choosing the right problem is everything. When you work on a real, painful challenge for a real market, he says, customer feedback is immediate — and that feedback loop is what makes agentic AI both powerful and practical.
Building AI-Native Applications to Enable Rapid Innovation
Kuo also outlines how his organization is approaching AI transformation internally, describing it as a three-layer strategy that reshapes both product development and the way teams operate:
- Creating AI-native applications that exist outside the constraints of existing business
- Integrating AI directly into the existing platform
- Giving every employee explicit AI KPIs
In the first layer, he discusses creating AI-native applications that operate outside the constraints of the existing business.
Their new platform, Accio, was built according to Kuo’s approach. Yet he notes that, because it’s separate from Alibaba.com’s legacy structure, his team can abandon old assumptions, redesign workflows from scratch, and move quickly without worrying about revenue targets or operational pressures tied to the core business.
For Kuo, this freedom enables them to experiment with entirely new models and build disruptive products at speed.
The second layer focuses on AI and Alibaba.com, where AI is integrated directly into the existing platform. Here, he notes they are redesigning core functions like search, recommendations, seller tools, advertising systems, and even deploying AI agents that take over as much as 80% of the routine operational work typically handled by staff.
Kuo is pleased to report that this layer is already showing tangible business impact: AI-driven features now contribute 10% of Alibaba.com’s annual revenue. That clear connection to value, he suggests, is the most evident proof that AI is not merely an enhancement but a growth engine within established operations.
The third layer Kuo describes centers on AI inside the organization itself. Every employee, from developers and operations teams to finance, risk, and compliance, now has explicit AI KPIs. Each person must identify how AI will transform their current workflows, improve efficiency, or automate tasks.
He highlights these human-oriented enterprise capabilities as the best kind of cultural shift — one where AI isn’t confined to dedicated teams but becomes a responsibility distributed across the entire enterprise.
Source link
#Scaling #Global #Trade #AIPowered #Tools #SMBs #Kuo #Zhang #Alibaba.com
