(Yicai) May 23 -- Chinese banks rushed to embrace artificial intelligence but have been disappointed by the results despite making massive investments, leading many to rethink and refine their AI strategies toward more specialized, scenario-tailored solutions, according to people in the banking and financial technology sectors.

Financial institutions use AI mainly to boost operational efficiency and cut costs, but it has yet to revolutionize their business models, said Fang Yue, director of the China Europe International Business School’s research center for AI and management innovation.

This is because the enterprise-level use of generative AI is still in its early stages, and its actual capabilities in areas such as smart investment research, risk control, and claims processing are still limited, Fang noted.

A fintech professional at one joint-stock bank said investment suggestions generated by large language models often deviate significantly from market realities, leading to lower-than-expected returns. They also lack the sophistication to detect new forms of financial fraud and sometimes even flag normal business communications as risk events, the source added.

Financial knowledge makes up only about 5 percent of pre-training data in foundational models, limiting their depth of understanding in finance and hindering deeper industry application, according to Meng Qian, chief information officer of the Bank of China.

Cost is another hurdle. LLMs require substantial capital and manpower from purchase and deployment to maintenance and optimization, which is a particularly burden for small and mid-sized financial firms, said Hu Debin, vice president at Bank of Shanghai. Even after these investments are made, it is often difficult to fully realize the model’s potential, Hu pointed out.

Fang at CEIBS said the notion of solving every problem with one monolithic model is clearly flawed., adding that big banks should avoid blindly adopting off-the-shelf LLMs and focus on building smaller, purpose-built models tailored to their unique business needs.

Zhao Liming, director of the Digital Finance Center at Kunming-based Fudian Bank, echoed this view. He suggested that small- and mid-sized banks leverage local data and internal resources to fine-tune their own models for better outcomes.

To this end, lenders need to hire tech talent with banking experience. When hiring AI professionals, banks should prioritize candidates with a strong understanding of corporate environments and business processes -- not just technical skills -- to better integrate AI into existing systems, Fang concluded.

Editors: Tang Shihua, Emmi Laine