Zhou began by revealing that roughly 40% of his presentation slides were generated by his company's proprietary AI office platform, with him personally refining the remaining 60%. He explained the title "From Expertise to Intelligence," noting that while traditional banking relied on human experts for system development, risk control, and model building over the past 20-30 years, large AI models now offer tangible methods to embed this expert knowledge directly into intelligent systems. This enables a broader range of users, from customers to internal staff and relationship managers, to access a degree of expert-level capability.
The company, which is dual-listed in the US and Hong Kong, established its internal AI large model department three years ago. Beginning this year, it started constructing its internal AI office capabilities, leading to a comprehensive "one horizontal, multiple vertical" full-stack solution. The horizontal layer is designed for banks and financial institutions, featuring an enterprise-level AI office platform named Qwork. The verticals focus on finance-specific scenarios, creating specialized agents in two primary areas.
The first area is credit and risk management. Here, they developed multi-agents including an AI due diligence officer for relationship and risk managers, an AI approver for review and approval staff, as well as an AI modeler, AI anti-fraud specialist, and AI strategist for risk management teams. This addresses a critical challenge for small and medium-sized banks, many of which lack quantitative risk management experts. A surprising number of banks with assets under 500 billion yuan have fewer than five people skilled in strategy modeling, often relying on external tech companies with varying reliability. The AI offers a new solution to this long-standing problem.
The second domain is marketing growth. Here, they built an AI loan officer for credit staff, an AI wealth advisor for wealth managers, an AI supervisor for performance management at branch and manager levels, and an AI operator to automate repetitive tasks. An AI growth officer acts as a central command hub for business growth within head office departments. Additional agents cover compliance and intelligent investment, creating a comprehensive matrix across both horizontal and vertical dimensions.
Detailing the Qwork platform, Zhou explained that it powers the entire company's internal operations. The AI integrates the company's various systems, data, documents, and knowledge, including analytical reports. This includes email, OA, an internal instant messaging system similar to NVIDIA's management approach, and an extensive cloud document system containing over 200,000 documents with built-in permission controls. The AI connects to production systems like data platforms, risk control, operations, and marketing, as well as backend systems including HR, finance, and administration.
Several critical components underpin the platform's security and effectiveness. Permission management is crucial, ensuring the AI can only access data the user is authorized to see. They employ a foundational model routing system, having deployed DeepSeek V4 locally while also leveraging public cloud models from Zhipu, Kimi, and GPT. An agent platform routing system pairs tasks with the most suitable coding tools. Finally, an AI security gateway determines whether tasks can use domestic or international public cloud models or must stay on local infrastructure, classifying data sensitivity based on keywords, MCP types, and URLs.
This setup dramatically lowers the barrier for all employees to safely access and use the world's best AI models. They integrated Qwork into their internal collaboration tools, and it can also be embedded into platforms like WeChat Work and DingTalk. The platform features expert skills, connectors using MCP protocols, and access to multiple base models in auto or specified modes. An internal skills marketplace allows employees to share and reuse custom-developed skills.
Zhou shared two compelling examples. First, he asked the AI to create his presentation by drawing from all his authorized cloud documents, past forum speeches, and local files. While the task took about half an hour, the AI produced a presentation of surprising specificity, far more detailed than generic output. Second, he asked the AI to analyze all his emails, Teams chats, and the T5T system over the past week to identify the most valuable bank client and the one with the most service issues. The AI correctly identified Guangdong Huaxing Bank as the most valuable client, providing detailed reasoning for its conclusion.
Focusing on specific agents, the AI strategist revolutionizes risk strategy development. Its six-step process includes infrastructure setup via MCP, building a knowledge base from past experts' documented experiences, data cleaning to understand variables, automatic strategy rule generation, deployment and validation with decision engines, and continuous monitoring. This transforms strategy iteration from a monthly process to an hourly one, makes it accessible to new graduates, uncovers hidden feature variables, and ensures institutional knowledge remains even if expert employees leave.
The AI anti-fraud specialist addresses the challenge of investigating suspicious defaults. Previously, a thorough case review could take a week or a month, requiring deep knowledge across multiple systems. Now, the AI automatically gathers evidence from diverse systems in minutes, consolidates expert insights, connects to external data sources like Qichacha for negative information, and identifies related fraud groups. While the AI proposes new anti-fraud rules, a human team ultimately makes the final decision on which candidates to implement.
Zhou concluded by highlighting Qifu Digital Technology, founded in 2019, is dedicated to advancing the digital and intelligent transformation of the banking sector. With substantial talent resources and extensive practical experience, its intelligent agent matrix is currently being implemented across various specialized fields. He expressed eagerness to further collaborate with industry experts and peers to drive this transformation forward.