SMEs are overwhelmed by multi-step operational work that requires time and staff. Global business processes remain slow, fragmented and opaque.
For all the focus on large companies in AI adoption, documentation from the World Trade Organization task force Remarks that SMEs represent almost 90-95% of all global businesses and account for 60% of global employment. Findings from other NGOs show that they continue to face outsized obstacles in trade and finance, which directly correspond to new heights of AI adoption in these sectors..
The World Economic Forum reports that small businesses particularly face issues that also amplify a “polycrisis” of disruptive economic shocks to the supply chain and other concomitant global events: limited access to capital, complex regulatory requirements, inadequate digital infrastructure, and difficulty adapting to the rules of international trade as smaller entities.
Numerous reports highlight the serious financial challenges facing SMEs. The Federal Reserve’s 2025 Small Business Credit Survey Remarks 75% cite rising costs of goods, services and wages as the main problem, with 56% struggling to pay operating expenses and 51% facing inconsistent cash flow. Funding needs led 59% to seek new credits, mainly for operations (56%), but 24% received none and 36% received only partial approval.
But financial difficulties are only part of the challenge. Even as SMEs struggle with rising costs and restricted access to capital, they also struggle to adopt the very technologies that could ease these burdens, particularly AI.
By Workday SMB Group and 2025 AI Trends report, SMBs face significant barriers to adoption: 48% say they don’t see clear business relevance, 47% cite security and privacy concerns, 28% highlight high costs, and 25% say they lack the skills or resources needed to implement AI.
In a recent episode of AI in Business, Emerj Editorial Director Matthew DeMello spoke with Kuo Zhang, Chairman of Alibaba.com, following his speech at this year’s conference. CoCreate 2025 event in Las Vegas, to discuss how SMBs and large enterprises can adopt agentic AI.
Their conversation highlights two critical insights for SMBs using AI that this article will explore in depth for retail, technology, and SMB leaders:
- End-to-end sourcing and purchasing automation for SMBs: By leveraging agentic AI as a force multiplier, SMBs can automate complex, multi-step operations that once required entire teams.
- Selecting the correct issues as a base layer: Apply a three-tier methodology of problem selection, leverage data and domain knowledge, and deploy models to create effective agentic AI for SMEs.
- Create native AI applications to enable rapid innovation: Native AI platforms enable SMBs to rethink their workflows, experiment with new models, and accelerate product development without legacy constraints.
Listen to the full episode below:
Guest: Kuo ZhangPresident, Alibaba.com
Skill: SMEs, Artificial Intelligence
Brief recognition: Kuo has been with Alibaba Group since 2011, where he previously led the merchant services business unit, overseeing operational tools, platform development and ecosystem management. In his current role, his objective remains consistent with the company’s core mission: facilitating the activities of SMEs, wherever they are. He holds a master’s degree in high performance computing from Tsinghua University, China.
End-to-end Sourcing and Procurement Automation for SMBs
Kuo opens the conversation by explaining that the company’s vision for agentic AI goes far beyond assistants generating documents or planning trips. Their platform, Acciolaunched last year, is now used by more than two million SMEs. It was designed to support real operational work that typically requires large teams and extended deadlines.
He illustrates this with a regional sporting event in Latin America that required the purchase of hundreds of thousands of items for a multi-country competition. A process that once took four months – identifying suppliers, checking compliance rules and coordinating requests – now takes hours.
Users upload an Excel sheet with items and requirements, and Accio interprets the details, checks import regulations, analyzes Alibaba’s network of 200,000 suppliers, generates a list of verified suppliers and even launches outreach until a final decision is ready.
He also shares a second example involving a solo entrepreneur who designs clothing for children with ADHD. What would typically require weeks of market research, design iterations, specification writing, vendor outreach, and prototyping discussions can now be executed by the agent system.
Kuo describes to the Emerj podcast audience how Accio takes the initial idea, researches the market, drafts detailed designs and specifications, and contacts appropriate manufacturers – something that previously required a lot of time and coordination.
It stands in stark contrast to traditional search engines, where users enter keywords and sort through thousands of results on their own. In contrast, the agentic model allows SMEs to upload full context: designs, specifications, product listings or complete spreadsheets. The system produces an execution plan that users can edit in plain language, then automatically launches and manages the entire workflow.
Kuo emphasizes that the experience is completely turnkey: no installation, no configuration, and instant use via a browser. For Alibaba.com, the simplicity he describes is at the heart of the vision: a plug-and-play agent replaces manual, tedious work with an autonomous system capable of running end-to-end business processes.
Select the right issues as a base layer
Kuo lays out the philosophy behind how his team creates agentic AI tools, presenting it as a three-tiered methodology that starts well before a model is trained:
- Choose the right problem
- Leverage data, domain knowledge and operational history
- Hardening with underlying models and infrastructure
The first and most essential step, he says, is to choose 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 actually felt by users:
“In our case, the main challenge we face is the complexity of global trade. This is a very real and important problem, representing approximately $30 trillion in global economic activity.”
Given its scale, it is a challenge worth taking up. With over 26 years of experience, we have developed a deep understanding of global buyers and suppliers and the challenges they face in the markets.
–Kuo Zhang, president of Alibaba.com
The second layer Kuo describes is the ability to leverage the data, domain knowledge, and operational history the company already has. 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 emphasizes that this foundation is important because agentic AI must not only generate a result – that result must be correct, reliable and operational. This requires careful design, context awareness, and an assessment framework grounded in real-world activity.
The third layer is the underlying models and technical infrastructure. Kuo notes that the company takes some of the world’s most important language models and layers its own applications and domain-specific insights on top of them. But even here he emphasizes that the model is not the starting point; it’s the final ingredient in a system built on problem selection, domain expertise, and rich data.
Through these levels, Kuo returns to a central idea: choosing the right problem is essential. When you’re working 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.
Create native AI applications to enable rapid innovation
Kuo also explains how his organization approaches AI transformation internally, describing it as a three-tiered strategy that reshapes both product development and the way teams operate:
- Create native AI applications that exist outside of existing business constraints
- Integrate AI directly into the existing platform
- Give each employee explicit AI KPIs
At the first level, it discusses creating native AI applications that operate outside of the constraints of the existing business.
Their new platform, Accio, was built using Kuo’s approach. Yet he notes that because it is distinct from Alibaba.com’s existing structure, his team can abandon old assumptions, rethink workflows from the ground up, and move quickly without worrying about revenue goals or core business operational pressures.
For Kuo, this freedom allows him to experiment with entirely new designs and quickly create revolutionary products.
The second layer focuses on AI and Alibaba.com, where AI is integrated directly into the existing platform. Here, he notes that they are rethinking core functions like search, recommendations, sales tools, advertising systems and even deploying AI agents that take over up to 80% of the routine operational work typically done by staff.
Kuo is happy to report that this layer is already having a tangible business impact: AI-based features now contribute 10% of Alibaba.com’s annual revenue. This clear link to value, he suggests, is the clearest evidence that AI is not just an improvement but an engine of growth within established operations.
The third level, Kuo, describes the AI centers within 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 emphasizes that these human-oriented enterprise capabilities are the best kind of cultural change: one in which AI is not limited to dedicated teams but becomes a distributed responsibility across the entire enterprise.