During the Spring Festival, the launch of the domestic large model DeepSeek sent shockwaves through the tech industry.

This has also stirred up ripples in the financial sector, particularly the banking industry, which boasts vast amounts of data and diverse scenarios. Many banks have taken the initiative to "tackle the unknown" by deploying the DeepSeek large model in various applications such as intelligent contract management, risk control, asset custody and valuation reconciliation, and customer service assistance.

It is noteworthy that previously, due to high costs and technical barriers, developing large models was often the domain of large banks. However, DeepSeek, an open-source, cost-effective, high-performance AI tool, appears to have presented small and medium-sized banks with an opportunity to bridge the R&D gap with their larger counterparts and further enhance the development of their own businesses.

01

Multiple Banks Embrace DeepSeek Empowerment

Suddenly, like the spring breeze that arrives overnight, the landscape changes dramatically.

Just days after DeepSeek went live and made a significant impact on the US stock market, numerous tech giants adopted a "proactive" strategy, including NVIDIA, AMD, Microsoft, Amazon Web Services (AWS), and Intel, which unanimously announced support for DeepSeek model services.

Domestically, the trend is no different. Leading cloud computing companies such as Huawei Cloud, Tencent Cloud, Tianyi Cloud, Alibaba Cloud, Baidu Intelligent Cloud, and Volcano Engine have also announced the launch of DeepSeek interfaces.

As a data-intensive industry, the banking sector was previously a priority for the implementation of large models. With the emergence of DeepSeek, many banks are intensifying their efforts to leverage its capabilities.

On February 11, according to the Tencent Cloud Intelligence official account, Chongqing Rural Commercial Bank announced on February 10 that, with the help of Tencent Cloud's large model knowledge engine capabilities, the bank has launched the intelligent assistant application "AI Xiaoyu" based on the DeepSeek model on Enterprise WeChat. This is also the first financial institution to build an online application based on DeepSeek through a knowledge engine.

The "AI Xiaoyu" intelligent assistant application will provide Chongqing Rural Commercial Bank's 15,000 employees with more efficient, convenient, and intelligent work support. By utilizing DeepSeek, the bank will also achieve breakthroughs in intelligent risk control, scenario finance, and data-driven decision-making.

Earlier news of DeepSeek deployment came from Bank of Jiangsu.

According to a message published on the official account "Jiangsu Financial Technology" on February 5, Bank of Jiangsu successfully deployed and fine-tuned the DeepSeek-VL2 multimodal model and the lightweight DeepSeek-R1 inference model, which are applied to intelligent contract quality inspection and automated valuation reconciliation scenarios, respectively. By mining and analyzing massive financial data, it has reshaped the financial service model, achieving dual breakthroughs in financial semantic understanding accuracy and business efficiency.

The official account also revealed that Bank of Jiangsu researched and developed the large language model service platform "Wisdom Xiaosu" in 2023. This time, by incorporating the DeepSeek large language model, "Wisdom Xiaosu" has been further enhanced in terms of processing capabilities for complex multimodal and multitasking scenarios, computing power savings, and efficiency.

Hainan Rural Commercial Bank, also located in Jiangsu, utilized DeepSeek for a marketing showcase. In an article titled "DeepSeek, You Understand Hainan Rural Commercial Bank Too Well!" published by the bank, it introduced the bank's situation to users by posing questions to DeepSeek. DeepSeek analyzed and summarized Hainan Rural Commercial Bank from multiple dimensions such as capital strength, market share, service quality, financial products, and social responsibility.

Sumer Bank is another notable example. Currently, the bank has begun to apply the DeepSeek VL2 multimodal model to process non-standard materials such as tables, image data, document images, etc., improving the comprehensive recognition accuracy of credit materials to 97% and integrating the DeepSeek R1 inference model into the self-developed "Development Assistant," shortening the iteration cycle of the core system by 30%. Additionally, Sumer Bank applies DeepSeek's distillation technology to over 20 scenarios such as credit risk control and anti-fraud, enhancing the efficiency of due diligence report generation by 40% and the accuracy of fraud risk labels by 35%.

Bank of Beijing, in collaboration with Huawei, is conducting pilot applications in multiple key business scenarios such as the AIB platform's Beijing Bank Research, Beijing Bank Think Tank, customer service assistant, and Beijing Bank Atlas.

Qi Finance has also learned that many small and medium-sized banks are quietly deploying DeepSeek or using it to develop their own large models.

02

Small and Medium-Sized Banks May Embrace New Opportunities

In fact, after the emergence of ChatGPT, over the past year or so, financial institutions, represented by state-owned and joint-stock banks, have explored the application potential of large models in numerous business scenarios through independent construction or joint development. In particular, in-depth practices have been carried out in intelligent customer service, intelligent investment advisors, digital employees, risk management, compliance reviews, and anti-fraud, covering all aspects of the front, middle, and back offices of banks.

So why are more and more banks embracing DeepSeek?

On the one hand, this is because DeepSeek offers better value for money. In the past, banks had to spend huge sums of money to purchase GPU equipment for self-developed large models, with a single training cost easily reaching millions of yuan. However, through algorithm optimization, DeepSeek maintains high performance while reducing training costs to one-tenth of traditional models.

On the other hand, unlike general large models that are "extensive but not precise," DeepSeek achieves vertical breakthroughs through a Mixture of Experts (MoE) architecture. For instance, in the intelligent contract quality inspection scenario, its multimodal model can simultaneously analyze text, seals, and signatures, mitigating the risk of misjudgment associated with a single modal model.

More importantly, banks can perform secondary development based on DeepSeek's open-source framework to quickly launch customized applications such as anti-fraud and asset valuation, significantly shortening the research and development cycle. This "Lego-style" innovation substantially lowers the technical threshold.

Under such circumstances, it is also expected to promote the integration of AI applications and realize AI inclusiveness. While large banks such as ICBC and CCB continue to increase their investment in self-developed models, small and medium-sized banks are shifting from relying on external large model service providers to the "open source + fine-tuning" model or forming "AI" alliances to share DeepSeek-derived models. This will narrow the technological gap between small and medium-sized banks and large banks, rewrite the AI competition rules where "large banks dominate," and enable them to compete on the same stage.

Additionally, in the process of digital transformation, especially the deployment of large financial models, many banks have not only transformed their business processes but also catalyzed deep changes in their organizational structures, breaking down traditional departmental barriers and further improving efficiency in human resources. The emergence of DeepSeek will further catalyze this transformation.

Although DeepSeek brings a ray of hope, it still cannot escape the common issues faced by large models, namely data security, privacy leaks, compliance challenges, and other risks. As Chief Researcher Dong Ximiao of China Merchants Union stated, "How to balance technological progress and risk prevention and control has become a key issue that financial institutions have to confront in the process of applying artificial intelligence technology."

A report from China Merchants Securities also indicates that although DeepSeek's reasoning ability has reached the international first-tier level, the issue of model hallucinations has not been completely resolved.

However, it is foreseeable that in an era of technological advancements, as the demand for data utilization security increases, large model applications that truly touch the core businesses of the banking industry will gradually become a reality.

The "fast and accurate" AI competition in the banking industry has already been ignited by DeepSeek.