Multi-step operational activity that consumes time and personnel overwhelms SMBs. International trade is still sluggish, dispersed, and unclear.
According to World Trade Organization working group literature, SMBs account for 60% of global employment and almost 90–95% of all global businesses, despite the concentration on major enterprises in AI deployment. According to other NGO research, they still encounter significant obstacles in the trade and financial sectors, which is directly related to the unprecedented levels of AI adoption in those domains.
According to the World Economic Forum, limited access to capital, complicated regulatory requirements, inadequate digital infrastructure, and the challenge of navigating international trade rules are among the issues that small businesses face most frequently. These issues also exacerbate a “polycrisis” of supply chain-shaking economic shocks and other concurrent global events.
The serious financial difficulties SMBs confront are highlighted in a number of papers. According to the 2025 Federal Reserve Small Business Credit Survey, 56% of businesses struggle to pay operational expenses, 51% experience inconsistent cash flows, and 75% mention rising costs of goods, services, and wages as the biggest problem. 59% of respondents sought new credit due to financing demands, mostly for operational purposes (56%). However, 24% received no credit, and 36% received only partial credit.
However, financial strain is just one aspect of the problem. SMBs struggle to embrace the very technologies that could alleviate these responsibilities, namely artificial intelligence (AI), even as they battle with growing expenses and limited access to funding.
SMBs have significant adoption challenges, according to SMB Group and Workday’s 2025 AI Trends report: 48% say they don’t see clear commercial relevance, 47% cite security and privacy issues, 28% point to high prices, and 25% say they lack the resources or skills necessary to apply AI.
Following his speech at this year’s CoCreate 2025 event in Las Vegas, Kuo Zhang, President of Alibaba.com, was interviewed by Editorial Director Matthew DeMello on a recent episode of AI in Business to talk about how both SMBs and businesses may use agentic AI.
Their discussion brings to light two crucial points for SMBs managing AI, which this piece will look at in-depth for executives in IT, retail, and SMBs:
- Automating SMBs’ whole sourcing and procurement process: SMBs may automate intricate, multi-step processes that formerly needed full teams by using agentic AI as a force multiplier.
- Choosing the appropriate issues to serve as the base layer: To create efficient agentic AI for SMBs, a three-layer methodology comprising problem selection, data and domain knowledge leveraging, and model deployment is used.
- Developing AI-native apps to facilitate quick innovation: Without being constrained by old systems, AI-native solutions enable SMBs to expedite product development, experiment with new models, and restructure workflows.
The entire episode can be heard below:
Visitor: Alibaba.com President Kuo Zhang
Knowledge of Artificial Intelligence and SMEs
Quick Acknowledgment: Kuo joined the Alibaba Group in 2011 and was formerly in charge of the Merchant Services Business Unit, which was in charge of platform development, ecosystem management, and operational tools. His focus in his present position is still in line with the company’s primary goal, which is to facilitate SMBs’ ability to conduct business wherever. He graduated from Tsinghua University in China with a Master’s degree in High Performance Computing.
Automating SMBs’ Complete Sourcing And Procurement Process
Kuo begins by outlining how the company’s goals for agentic AI extend well beyond assistants that create paperwork or arrange trips. More than two million SMBs are currently using their platform, Accio, which was introduced last year. It was created to replace actual operational work, which typically calls for sizable teams and lengthy schedules.
He used a Latin American regional sporting event that required the procurement of hundreds of thousands of products for a multi-nation competition to demonstrate this point. Finding suppliers, verifying compliance regulations, and coordinating inquiries used to take four months, but now it takes hours.
Accio analyzes the information, verifies import laws, searches Alibaba’s 200,000-supplier network, creates a verified supplier list, and even starts outreach until a final decision is made. Users provide an Excel sheet with the items and requirements.
He also gives a second example of a lone proprietor creating clothes for kids with ADHD. The agentic system can now carry out tasks that would typically take weeks of market research, design iterations, specification preparation, supplier outreach, and prototyping talks.
Kuo explains to the listeners of the podcast how Accio evaluates the original concept, does market research, writes drawings and comprehensive specifications, and contacts appropriate manufacturers—all of which previously required a substantial amount of effort and coordination.
He contrasts it sharply with conventional search engines, where users manually sort through thousands of results after entering keywords. The agentic architecture, on the other hand, enables SMBs to submit comprehensive context, such as product lists, designs, specifications, or entire spreadsheets. The technology automatically initiates and oversees the entire workflow after producing an execution plan that users can modify in plain English.
Kuo emphasizes that the experience is completely turnkey, requiring no setup or installation and instant browser use. The simplicity he outlines is essential to Alibaba.com’s vision: an innovative agent substitutes laborious, manual labor with an autonomous system that can manage end-to-end commercial operations.
Choosing the Appropriate Issues For The Foundation Layer
Kuo describes the three-layer approach that his team uses to develop agentic AI tools, which begins long before any models are trained:
- Selecting the appropriate issue
- Making use of data, domain expertise, and past operations
- Using infrastructure and underlying models to reinforce
- He claims that selecting the appropriate problem is the first and most crucial layer. He maintains that many products fail because the issue they aim to solve is either unimportant or not really perceived by people, rather than because the technology is flawed:
The intricacy of international trade is the main issue we are tackling in our situation. This issue, which accounts for over $30 trillion in worldwide economic activity, is quite real and important.
It is a challenge worth taking on because of its size. With over 26 years of experience, we have a thorough understanding of international suppliers and buyers as well as the challenges they encounter in various markets.
—Kuo Zhang, Alibaba.com President
Utilizing the data, subject expertise, and operational history that the business already has is the second layer that Kuo outlines. The system can learn what a successful response looks like in context thanks to the billions of products, hundreds of millions of customers, hundreds of thousands of suppliers, and the steady stream of daily transactions.
He emphasizes the importance of this foundation since agentic AI needs to produce an output that is accurate, dependable, and operationally sound. This calls for thoughtful planning, contextual knowledge, and an assessment system based on practical experience.
The technical infrastructure and underlying models are part of the third layer. According to Kuo, the business builds its own apps and domain-specific intelligence on top of some of the biggest language models in the world. Even yet, he stresses that the model is the last component in a system that is based on rich data, domain expertise, and issue selection.
Kuo keeps coming back to one main point throughout all of these layers: picking the correct challenge is crucial. He claims that client input is instantaneous when working on a genuine, difficult problem for a real market, and that feedback loop is what makes agentic AI both effective and useful.
Developing AI-Native Applications To Facilitate Quick Innovation
Additionally, Kuo describes the internal AI transformation method his company is using, characterizing it as a three-layer approach that transforms both product development and team dynamics:
- Developing AI-native solutions that are not limited by current business
- Directly incorporating AI into the current platform
- Providing clear AI KPIs to each staff
- He talks about developing AI-native applications that function outside of the limitations of the current business at the first layer.
Accio, their new platform, was developed using Kuo’s methodology. However, he points out that because it is distinct from Alibaba.com’s legacy structure, his team can quickly move forward, rebuild workflows from the ground up, and let go of outdated assumptions without being constrained by operational pressures or revenue targets related to the core business.
This flexibility allows Kuo to quickly develop game-changing goods and test whole new concepts.
The second tier focuses on artificial intelligence (AI) and Alibaba.com, where AI is directly incorporated into the platform. Here, he points out that they are rethinking essential features like search, recommendations, seller tools, advertising systems, and even implementing AI bots that perform up to 80% of the regular operational tasks normally performed by employees.
Kuo is happy to say that this layer is already having a noticeable effect on business: 10% of Alibaba.com’s yearly income currently comes from AI-driven features. He claims that this obvious link to value is the best evidence that AI is a growth engine within established processes rather than just an improvement.
Kuo focuses on AI within the company itself in the third layer. Every worker now has clear AI KPIs, from the operations and development teams to the finance, risk, and compliance departments. Each individual must determine how AI will automate activities, increase productivity, or change their current workflows.
He emphasizes these human-centered enterprise capabilities as the ideal form of cultural change, where AI isn’t limited to specialized teams but rather becomes a shared responsibility throughout the entire company.