DeepSeek's One-Year Anniversary: Has the Large Model Changed Commercial Banks? Significant Cost Reduction in Marketing, More Effects to Be Observed

Last January, DeepSeek made its debut, marking the beginning of the domestic banking industry’s wave of embracing large models. Now, a year later, what changes have large models actually brought to banks? Over the past two weeks, Caixin reporters have conducted in-depth exchanges and interviews with industry insiders.

Several bank officials told reporters that, from actual application results, large models are already playing a significant role in areas such as marketing, intelligent customer service, and staff training. They have some effect on risk control optimization, but further observation is needed. Regarding claims from some banks that large models can enhance core risk identification capabilities or even “reshape” the risk management system, multiple industry insiders advised against excessive optimism, emphasizing the need for more time and case comparisons.

“DeepSeek only debuted last year, and some banks said in the second half of the year that it helped optimize risk control, but I find that hard to believe,” pointed out a city commercial bank official. He noted that the bank’s own risk management system is already relatively mature and automated, with human intervention greatly reduced. The key lies in the authenticity of the source documents submitted for review, i.e., the credibility of data sources. While large models can indeed speed up review efficiency, whether they effectively improve risk control accuracy requires at least a year or longer of comparative analysis before conclusions can be drawn.

Large models save banks significant marketing costs, but more effects need to be observed

An official from a listed bank told reporters that due to marketing and brand promotion needs, the bank’s annual expenditure on designing marketing posters and releasing brand advertisements has generally been around tens of millions. These tasks were previously outsourced to other companies, which handled them after the bank’s requests. “Many times, we were not satisfied and had to revise repeatedly.”

However, after DeepSeek’s debut last year, the bank guided its staff in AI design applications, achieving very good results. “Although human intervention is still needed, considering other costs, last year’s expenses for this dropped to around one million.” He explained that reducing costs from 10 million to 1 million shows that AI large models have indeed brought tangible benefits to the bank in certain areas.

“After adopting large models, we naturally no longer need so many outsourcing companies,” the official from the listed bank pointed out. He believes that once AI large models mature internally, some offline positions may gradually be reduced or even eliminated.

In staff training and intelligent customer service, AI large models have already achieved steady application in banks. Industry insiders explained that, in the past, training for grassroots staff, especially financial advisors, required significant time and manpower from headquarters. But this has changed with AI intervention. Currently, the headquarters is exploring ways to transmit AI-optimized marketing scripts and related financial knowledge to grassroots staff through backend systems, enabling them to better sell products with the help of AI models.

A fintech staff member from a joint-stock bank pointed out that the greatest value of AI large models currently may lie in their “transfer learning” ability, which allows summarizing and generalizing experience from mature scenarios and then expanding to similar new scenarios. Therefore, they already have certain applications and advantages in marketing financial products.

However, during interviews and exchanges with industry insiders, the positive contributions of large models seem to be limited to this point. Further application in business areas remains mostly experimental and controversial.

The reporter learned that, aside from risk control, the currently popular field of intelligent marketing still faces disputes over AI application. A senior official from a listed bank’s headquarters said that with advances in AI technology, the ideal future scenario might be: AI robots providing investment advice, eliminating the need for financial advisors. However, this view is not widely accepted among peers. One banker told the reporter, “AI marketing that makes everyone look the same may not suit everyone. I personally prefer genuine human interaction.”

No killer app yet, more competitors entering the bank large model arena

A staff member from a listed city commercial bank told the reporter that in February last year, after DeepSeek’s debut, the bank also introduced the large model and tasked technical staff with deep learning, deployment, and R&D. Unfortunately, the bank has yet to develop a sufficiently mature and useful application based on DeepSeek. Overall, DeepSeek’s application in banking has not demonstrated enough advantages. “It’s no longer fresh.” Although the bank still deploys DeepSeek, “expectations have lowered,” and they do not anticipate a disruptive or killer app emerging soon.

Another local city commercial bank official also confirmed that last summer, after realizing DeepSeek’s limitations in financial data, they attempted to integrate their existing single-business “small models” with DeepSeek to develop practical applications. But so far, results have been limited, and further R&D and refinement are needed.

Caixin reporters learned that the current biggest role of DeepSeek may be to stimulate more industry peers to follow quickly and compete fiercely in fields like finance and healthcare. Since mid-last year, large models including Alibaba’s Qianwen have been exploring and collaborating with multiple banks, achieving positive results. Previously, a leading state-owned bank announced a partnership with Alibaba.

“Alibaba and our bank recently also collaborated. Our bank has also purchased and deployed Alibaba’s large model,” confirmed a senior official from a leading city commercial bank. He said that the bank is adopting a “multi-pronged” strategy. However, related applications are still in exploration, as banks have higher requirements for data security and system stability.

(Article source: Caixin)