MicroPort

1. Necessity of AI Adoption

AI has become an inevitable choice for enterprise development. Taking the example of the

DeepSeek model, which became widely popular overnight during the 2025 Spring Festival,

Chinese enterprises have realized the urgency of AI transformation. As a core technology of the

Fourth Industrial Revolution, AI will profoundly change the business landscape. Enterprises should actively embrace AI to avoid being eliminated by technology.

2. Concerns of Manufacturing Enterprises

China has the largest number of manufacturing enterprises in the world. However, most of these enterprises still have a limited understanding of artificial intelligence technology. In the process of adopting AI, these are the common concerns many enterprises have:

1. The AI Black Box: The training and inference process of large models is not transparent.

For non-AI enterprises, AI remains a black box. Even though there are techniques like prompt engineering and fine-tuning, the results still have uncertainty.

2. Continuous Optimization: AI models are constantly evolving. Enterprises lack effective evaluation mechanisms to introduce, monitor, and optimize AI models.

3. Practical Effectiveness: It is crucial to ensure that AI truly helps manufacturing enterprises improve quality, reduce costs, and increase efficiency. AI models need to meet the requirements of accuracy, explainability, and stability while minimizing AI hallucinations.

4. Data Security for Enterprise Knowledge: Enterprise core knowledge could be memorized by AI models, leading to potential leakage risks. If AI is integrated into core systems like

ERP or CRM, protecting operational data becomes critical.

5. Talent Gap: Traditional enterprises often lack professionals who can effectively use and maintain AI technologies, especially in areas such as local deployment and model fine-tuning.

6. Organizational Change: Employee training and cultural transformation are necessary to ensure that staff understand and adopt AI, facilitating a smooth AI deployment.

“Companies should assess AI applications based on accuracy, stability, interpretability, and tangible business impact rather than simply treating AI as an experimental or showcase technology”

3. AI Model Selection

Selecting the right AI model is crucial for manufacturing enterprises, as different business needs require different approaches. For scenarios focused on improving individual productivity, companies can leverage internet-based AI models, such as intelligent assistants and automated document processing tools. However, when AI is applied to core business operations, locally deployed open-source models are often the better choice, ensuring data security and greater control.

In certain applications, businesses require a high level of accuracy from AI, such as in production optimization and supply chain forecasting. In these cases, fine-tunable open-source models are preferable, as they can be adapted to specific business requirements, improving relevance and precision. Additionally, a multi-model strategy can be beneficial, where different models are used for different scenarios, and cross-validation techniques are employed to enhance overall reliability.

4. Choosing the Right AI Application Scenarios

When integrating AI into operations, enterprises must carefully evaluate the true value of each application and develop a structured implementation strategy. Many companies are eager to explore AI, with various departments looking for ways to leverage the technology. However, without proper assessment criteria, businesses may end up investing in projects that offer limited practical benefits.

A practical approach is to prioritize common-use scenarios, such as improving employee efficiency and streamlining internal processes. These applications can be quickly implemented while ensuring Cybersecurity and fostering AI literacy among employees. At the same time, businesses should focus on high-value applications, particularly those involving complex business logic in specialized vertical areas. Unlike general-use cases, these scenarios often require industrial or enterprise-customized AI models, making it essential for enterprises to build the necessary in-house technical capabilities (as they might not have the right vendor) and accumulate expertise in AI governance.

Moreover, establishing clear evaluation standards is key to ensuring AI projects generate real business value. Companies should assess AI applications based on accuracy, stability, interpretability, and tangible business impact rather than simply treating AI as an experimental or showcase technology.

5. AI Application in MicroPort

MicroPort has implemented AI in several pilot use cases. One of our primary applications is an intelligent Product Knowledge Chatbot that assists sales teams and dealerships in quickly accessing product knowledge and materials. This not only improves internal training efficiency but also enhances the accuracy and consistency of external communication.

Secondly, AI has been deployed in knowledge management. There are unstructured documents scattered across systems, cloud drives, and personal devices; we use AI technology to read unstructured documents, generate tags, identify linkages and curate knowledge into knowledge graphs. This approach improves knowledge accessibility while reducing communication gaps caused by information silos. We also think this is a critical step for adopting AI applications because they all rely on high-quality data.

In IT development, we use AI-assisted coding tools to optimize software development processes, leading to increased efficiency and better system stability.

Most recently, we have launched an AI-driven data analytics system - MIND. MIND system collects both public information in medical device industry, and enterprise internal sales and marketing related data automatically. With ChatBI (conversational business intelligence) and workflow automation using AI Agent, it allows management team to access valuable insights more quickly and refine their strategies effectively.

6. Key Takeaways

We have identified several key insights from our experience as key takeaways.

1. Enterprises must establish robust AI model evaluation capabilities to ensure that the technology aligns with their strategies.

2. AI scenario selection must be strategic. Companies should prioritize applications that address common challenges and offer high-impact benefits, such as efficiency improvements and process automation, while also exploring innovative use cases that align closely with their long-term business strategies.

3. Effective data management is another critical factor, as high-quality input data directly influences model performance. We have also found that integrating AI with knowledge graphs and prompt engineering helps reduce hallucinations, making models more reliable.

4. Investing in AI talent is essential, particularly for areas such as local deployment and model fine-tuning, where having an in-house team significantly increases the success rate of AI adoption.

5. Finally, security remains a top concern for enterprises. When using commercial AI models, businesses must take extra precautions to protect sensitive data and prevent core business information from being compromised.

AI is reshaping the business landscape for manufacturing enterprises, from process automation to intelligent decision-making. While the potential of AI is immense, companies must approach its adoption with a well-defined strategy and a rational mindset. By aligning AI with business goals and continuously refining its applications, enterprises can achieve deep integration of AI and business operations, driving long-term value and competitive advantage.