At the "2025 Greater Bay Area Exchange Technology Conference" jointly hosted by the Shenzhen Stock Exchange, Hong Kong Exchanges and Clearing, and Guangzhou Futures Exchange on November 28, Han Zhencong, Chief Information Officer of HTSC, delivered a keynote speech highlighting that AI technological innovation has become a critical pathway for the financial industry's development. The sector is accelerating AI's penetration from peripheral support into core financial services.
Global leading financial institutions are spearheading efforts to integrate AI into core scenarios. Overseas financial institutions have demonstrated the trend of deep AI applications in finance. Currently, AI assistants with high penetration rates cover a significant portion of financial professionals, improving code generation and testing efficiency by over 40%, while increasingly embedding into core areas such as trading systems and risk control models. These cases clearly show that top financial institutions are no longer satisfied with AI applications in peripheral tasks like email drafting and document summarization but are firmly advancing into core domains such as trading and investment advisory.
Despite the AI deployment boom, financial institutions remain cautious in core business scenarios. Data reveals that current AI applications in finance predominantly focus on non-decision-making areas like knowledge management (49%), accounts payable automation (37%), and error/exception detection (34%). In contrast, adoption remains low in high-stakes core processes such as trading decisions and risk pricing.
Two major barriers hinder AI's breakthrough in core scenarios. First, the "cost of hallucinations": general AI's limited accuracy may introduce systemic risks in financial trading and risk control. OpenAI's October 2025 policy update explicitly prohibits ChatGPT from providing financial advice, underscoring the inherent limitations of general AI in core financial applications. Second, the "fuel gap": while tech companies possess cutting-edge algorithms, they lack two critical components—high-quality, sensitive internal financial data restricted by privacy regulations and deep industry insights into market logic and economic cycles. Financial sector-specific "Know-How" cannot be acquired through algorithmic training alone.
Financial institutions must lead innovation in core scenarios. A clear division of labor has emerged: tech companies focus on infrastructure and general applications, while financial institutions must drive breakthroughs in value-critical scenarios. HTSC exemplifies this approach with its "All in AI" corporate strategy, aiming to deeply integrate financial logic with AI.
HTSC prioritizes overcoming the "data and cognition" barriers. Recognizing that critical data and expertise are often siloed across departments and systems, HTSC launched systematic integration initiatives. It employs technical solutions for cross-departmental data consolidation while incorporating standardized inputs from industry experts and senior analysts—capturing their judgments on trends and valuations into models.
For model architecture, HTSC adopts a "large model as teacher, small model as student" hierarchical approach. It trains tailored models for different business needs: specialized "small models" ensure millisecond response for high-stakes trading execution, while larger models handle complex macroeconomic and sector research. A rigorous data annotation and verification system guarantees training data accuracy.
Computing power forms the foundation. HTSC is building a "self-reliant, diversified" computing platform—deploying local high-performance GPU clusters for secure core operations while partnering with domestic computing leaders for flexible cloud resources. All computing investments align with practical needs across client services, investment research, trading, and risk management.
Showcasing its "All in AI" strategy, HTSC launched "AI Zhang Le" in October—a native AI application that redefines trading experiences. Unlike traditional apps with AI add-ons, it rebuilds user interaction from scratch, enabling natural language commands for stock screening and order execution. The product integrates HTSC's research expertise and compliance logic, leveraging real-time trusted data and proprietary search to ensure verifiable responses. Its annotation team comprises in-house experts and analysts, addressing general AI's reliability gaps in finance.
In the AI-native era, governance becomes paramount. Beyond traditional financial risks, model risks must take center stage. Given AI's potential for systemic errors and biases, the industry needs unified governance frameworks covering model lifecycle management, sensitive data isolation, training data anonymization/tracing, and continuous performance evaluation—ensuring all decision-making applications operate within secure, controllable boundaries. HTSC advocates collaborative development of AI standards and shared technological advancements to democratize intelligent finance under regulatory compliance.
Comments