Taishin Holding Transforms Customer Experience with FedGPT-Powered Knowledge Management

2025/05/09

“The aim was to break down information silos, empower frontline staff, and achieve intelligent knowledge management and efficient application, thereby enhancing overall service quality and customer satisfaction.”

Background

Company Name: Taishin Financial Holding | Industry: Financial Services

Challenges

As a leader in Taiwan's financial services industry, Taishin Financial Holding began exploring how to leverage Large Language Models (LLMs) to solve several pain points during the generative AI explosion of 2024.

First, the company's vast professional knowledge and product information were scattered across different systems in various departments, making it difficult for employees to find the needed information quickly. This resulted in staff spending excessive time searching for data rather than serving customers.

Financial consultants in particular couldn't access the latest, complete product information in real-time during customer consultations. This not only extended customer waiting times but also reduced transaction success rates. The cumbersome traditional document search process forced consultants to spend several hours daily on information retrieval, preventing them from focusing on high-value customer relationship development.

Additionally, the customer service center faced high turnover challenges. With constantly evolving products and services, brand spirit and knowledge transfer became inefficient, affecting response speed and accuracy. The existing intelligent customer service built on Natural Language Processing (NLP) technology couldn't meet customer expectations for personalized experiences or effectively distribute the continuously growing workload of the customer service center.

Taishin Financial Holding urgently needed an integrated solution to break down information silos, empower frontline staff, achieve intelligent knowledge management and efficient application, thereby improving overall service quality and customer satisfaction.

Solutions

• Product: FedGPT: AutoKB, AutoML

Domain-Specific Fine-Tuning for a Financial-Grade AI Model

"X-Brain" deployed the FedGPT system locally, ensuring the security of critical industry data while providing Taishin Financial Holding with a controllable AI development path and reducing long-term operating costs.

Taishin's AI team collaborated with Yating Intelligence's algorithm experts to systematically organize Taishin Financial Holding's implicit and explicit knowledge as a pre-training dataset, aiming to make the model more aligned with business needs. After six months of optimization and fine-tuning, "X-Brain," deeply versed in core businesses like trust, wealth management, and credit, launched in Q1 2025, becoming the growth engine for Taishin Financial's various generative AI application services.

"X-Brain" uses AutoKB functionality combined with Retrieval-Augmented Generation (RAG) technology to integrate internal knowledge bases, ensuring AI responses are well-grounded and providing accurate, professional answers while effectively reducing AI hallucination risks.

Furthermore, FedGPT's AutoML capability leverages efficient QLoRA fine-tuning technology to help Taishin Financial Holding continuously optimize the model, allowing "X-Brain" to dynamically adjust to the latest business requirements. FedGPT not only reduces training costs but also ensures response accuracy and applicability, further strengthening intelligent knowledge management and customer service experiences.

Benefits & Results

AI-Empowered Business Development: Bringing Comprehensive Improvements

The three core business units under Taishin Financial Holding—banking, securities, and life insurance—injected their knowledge bases into "X-Brain," achieving these key outcomes:

• Intelligent Knowledge Management: Employees can query internal corporate data using natural language, gaining instant access to needed information and improving decision-making efficiency.

• Financial Consultant Performance Enhancement: Helps financial consultants quickly retrieve the latest product information, improving efficiency by 60% compared to traditional document search methods, allowing consultants to focus on customer relationship management and improving service quality.

• Customer Service Experience Optimization: Through RAG technology, ensuring response accuracy, customer service processing time has been reduced by 40%, significantly improving customer satisfaction.

In the future, "X-Brain" will continue to undergo model fine-tuning to further enhance its ability to adapt to different business scenarios, maximizing the value of AI and human-machine collaboration to create more competitive intelligent financial services for Taishin Bank.