As information technology rapidly advances, cloud computing, big data analytics, and artificial intelligence (AI) are no longer futuristic concepts. These emerging technologies have rapidly reshaped operational models across industries, enhancing corporate efficiency, competitive advantages, and fostering deeper customer relationships. Furthermore, they have catalyzed the development of new business models. As the demand for precise decision-making and effective problem-solving in professional fields continues to grow, large language models (LLMs) designed for specific industries have become widely adopted in areas such as healthcare, finance, and law. Unlike general-purpose language processing tools trained on extensive linguistic data, these specialized AI models integrate deep learning and domain knowledge, which enable higher adaptability and accuracy tailored to specific industry requirements.
Challenges and Applications of AI Integration
During a period of rapid AI innovation, TDCC launched the Exploration of Optimal AI Applications for Enhancing Digital Capabilities Project (hereafter referred to as “the Project”) in January 2024. Since the launch of the TDCC E-billing Platform in September 2022, serving nearly 2,500 financial institutions and issuing companies, most inquiries regarding service fees have surged, especially during monthly billing cycles. To address these inquiries swiftly and accurately, alongside routine operations, the Finance Department proposed integrating an AI-powered text-based customer service system to provide services that maintain a participant-centered approach and combine human warmth with technological convenience. To improve the system’s ability to respond effectively and accurately to participants’ inquiries, the department incorporated over 20 company service fee regulations into the training data. Our team also compiled and analyzed daily call inquiries to enrich the training dataset.
The Project employs retrieval-augmented generation (RAG) to train the AI on service fee data and frequently asked questions (FAQs) to construct an AI-powered text-based customer service system. RAG combines traditional retrieval techniques with generative AI capabilities, enabling the model to generate accurate responses based on pre-existing datasets. However, the initial training with over 20 service fee regulations produced disappointing results. A primary challenge was how to process large volumes of text data effectively, given the strict token limitations imposed by many language models. When the input text exceeds the model’s processing capacity, it leads to inefficiencies and inaccuracies and also restricts the model’s ability to deliver reliable outputs. In such cases, we couldn’t input the entire lengthy text into the model for analysis or processing, which significantly reduced the model’s comprehension of the overall data and even resulted in ineffective and inaccurate responses. To address this issue, with support from consultants and the Digital Development & Information Security Department, we resolved this issue by adopting chunking. This preprocessing method divides lengthy service fee regulations into smaller, relatively independent units (chunks). Each chunk focuses on specific paragraphs or topics, thereby overcoming token limitations while maintaining logical consistency within the text. This approach not only overcame token length limitations but also preserved the logical consistency within the document. It enhanced the AI model’s understanding of context and improved response accuracy, facilitating subsequent processing and analysis. Additionally, we categorized and analyzed nearly 300 service fee codes, refining the training data through iterative adjustments. Although occasional misinterpretations persisted with more complex queries, our continuous efforts in fine-tuning the model and expanding the training dataset led to gradual performance improvements. The AI-powered text-based customer service system now excels in handling routine service fees and FAQs. Testing results demonstrate its ability to respond accurately to the majority of inquiries, reducing the challenges faced by participants during monthly account reconciliations and garnering positive feedback that reflects increased customer satisfaction.
Future Outlook: Digitalization and Sustainability
Over the past few years, influenced by the pandemic, remote or hybrid work models has become the norm for most participants. To address this shift, the Finance Department has integrated ESG principles into its digital services, establishing the TDCC E-billing Platform. This platform consolidates all service fee information and online query functionalities, providing a fully digitalized query service. The introduction of the AI-powered text-based customer service system builds on this foundation. It aims to achieve a paperless environment and provide real-time responsiveness to customer needs. This advancement not only enhances service quality but also optimizes team efficiency, allowing staff to focus on more complex and challenging core tasks. By collaborating deeply with AI technology, we believe we can improve work efficiency while achieving shared growth and driving sustainable business development. In the near future, guided by the company’s proactive digital transformation strategy, the Finance Department remains committed to exploring innovative solutions. These efforts will streamline internal processes, elevate customer experiences, and ensure competitiveness in the digital age. Through deep collaboration with AI technology, we are confident in our ability to deliver efficient, intelligent, and human-centered services, promoting sustainable business development and maintaining a leading edge in the digital age.
