On September 10, the 2025 China International Fair for Trade in Services (CIFTIS) opened in Beijing, attracting nearly 2,000 companies from over 70 countries and regions to participate in this world's largest service trade event. During the 7th China Fintech Forum, Song Rongxin, Vice President of Technology at Qifu Technology, delivered a keynote speech titled "AI-Native + Finance: From Peripheral to Core," highlighting that current financial AI applications are rapidly penetrating from peripheral functions like customer service and document processing to core scenarios such as risk control and customer acquisition. AI-Native architecture serves as the key to achieving this transformation, with Qifu Technology's practices fully validating this trend and value.
Song Rongxin noted that globally, AI applications in the financial sector have demonstrated a clear evolutionary trajectory "from peripheral to core." According to Qifu Technology's research and analysis of over 600 financial AI products, approximately 85% of AI products in 2023 were still concentrated on peripheral functions such as customer service and document processing. However, by this year, one-third of products have entered core areas like risk control and customer acquisition. Taking Casca as an example, it has restructured bank lending platforms through large language model technology, achieving a 10-fold improvement in lending efficiency while reducing personnel investment by 90%, covering numerous small and micro enterprises that previously had insufficient financial services access, demonstrating the tremendous value of AI penetration into core financial scenarios.
Qifu Technology's own development path serves as a microcosm of this trend. Song Rongxin recalled that in 2023, Qifu assembled a hundred-person large language model team, with applications focused on relatively peripheral scenarios such as customer service assistants and marketing quality inspection. In 2024, AI assistants like Jarvis and ChatBI entered business efficiency enhancement areas but had not yet reached core functions. In 2025, with technological iteration and architectural maturity, intelligent agents based on AI-Native architecture achieved breakthroughs, with intelligent applications such as AI loan officers and AI approval officials entering core areas like risk control and customer acquisition, completing the critical leap from peripheral to core.
Song Rongxin emphasized that Qifu Technology's ability to achieve this breakthrough stems from three core pillars. First, long-term and sustained AI investment serves as the foundation. The company has invested nearly 10 billion yuan in R&D funding, maintains a technical team of nearly 1,000 people, has published multiple research papers at top international academic conferences, and led the development of China's first financial large language model standard, providing confidence for deep technological cultivation. Second, rich data and scenarios serve as the evolutionary soil. Daily processing of 180,000 application decisions, 370,000 transaction decisions, 600,000 linguistic samples, and over 16 million user interactions provide massive learning samples for intelligent agents. Taking customer manager reviews as an example, AI-Native intelligent agents can achieve accuracy rates exceeding 85% when combined with environmental data, far surpassing the 60% accuracy of manual processes and 75% of general large language models. Third, AI-Native architecture represents the key choice. Unlike traditional preset rule-driven approaches that only embed single processes, this architecture positions intelligent agents as system centers capable of processing multimodal information, possessing reasoning and learning capabilities, forming a "data-model-business" closed loop, and achieving continuous evolution.
Today, multiple innovative achievements demonstrate the deep application of AI in core financial scenarios. For example, AI loan officers serve as intelligent assistants to bank relationship managers, dynamically responding to customer needs and optimizing service processes, resulting in an average 15% improvement in conversion rates, 30% efficiency enhancement, and expansion of per-person customer management from hundreds to thousands. AI approval officials penetrate the most core financial approval and risk control processes, achieving zero supplementary document requirements, 100% automation, and T+0 processing time. In risk modeling, they can also serve as "senior modeling expert" assistants, continuously advancing work 24 hours a day.
Song Rongxin stated that the role transformation of AI in finance from peripheral assistance to core driving force is inevitable. China possesses the world's most complex financial scenarios and largest user base, providing an excellent foundation for deep AI evolution in finance. Qifu Technology will continue to focus on AI-Native architecture as its core, deeply cultivate core financial scenarios, and promote deeper integration between AI and finance, contributing Chinese solutions to global financial industry intelligent upgrades.