Around January last year, DeepSeek emerged, marking the beginning of domestic banking's embrace of the large model wave. Now, one year later, what changes have large models actually brought to banks? Over the past two weeks, in-depth exchanges and interviews were conducted with banking industry insiders. Representatives from multiple banks indicated that based on practical application results, large models have already demonstrated significant effects in areas such as marketing, intelligent customer service, and staff training. While they show some utility in risk control optimization, further observation is required. Regarding claims by some banks that large models can enhance core risk identification capabilities or even "reshape" risk control systems, several banking professionals expressed that excessive optimism is unwarranted, and more time and comparative case studies are needed for evaluation. "DeepSeek only launched last year; it's hardly credible for a bank to claim by the second half of the year that it helped optimize risk control," pointed out a city commercial bank representative. They noted that banks' proprietary risk control systems are already relatively mature and automated, with minimal manual intervention, and the key lies in the authenticity of source submission materials—the credibility of data sources. While large models can accelerate backend personnel review efficiency, determining whether they "effectively improve risk control precision" requires at least a year or longer of comparative analysis to draw conclusions.

Large models are saving banks substantial marketing expenses, but their broader utility remains under observation. A representative from a listed bank mentioned that due to marketing and brand promotion needs, the bank historically spent approximately ten million annually on tasks like designing marketing posters and releasing brand advertisements. These tasks were previously outsourced to external companies after the bank provided requirements. "Often, we weren't satisfied and required repeated revisions." However, following DeepSeek's launch last year, the bank guided its staff in applying AI design tools with excellent results. "Although human intervention is still needed, considering overall costs, this expense was reduced to one million last year." They stated that the drop from 10 million to 1 million demonstrates tangible benefits large AI models have already delivered in certain areas. "With the adoption of large models, our bank naturally no longer requires as many outsourcing firms," the listed bank representative noted. They believe that even within banks, as AI large models mature, some offline positions may gradually shrink or disappear. In staff training and intelligent customer service, AI large models have steadily gained application in banks. Relevant personnel explained that previously, training for frontline staff, especially wealth managers, required significant time and resources from the head office. AI intervention has changed this. The head office is now exploring transmitting AI-optimized marketing scripts and related financial knowledge to frontline bank staff via backend systems, enabling them to better perform product sales using AI models. A financial technology officer from a joint-stock bank indicated that currently, the greatest value of AI large models likely lies in their "transfer learning" capability—summarizing experiences from mature scenarios and extending them to similar new scenarios. Thus, they already possess certain applications and advantages in areas like wealth management product marketing.

However, based on interviews and exchanges within the industry, the positive contributions of large models largely end here. Further applications in business domains currently involve a mix of experimentation and controversy. It was learned that beyond the aforementioned risk control area, disputes persist regarding banks' use of AI in the currently popular field of intelligent marketing. A head office representative from a listed bank suggested that with advancing AI technology, an ideal future scenario might involve AI robots providing purchase recommendations for wealth management products, eliminating the need for wealth managers. This view, however, is not universally accepted. Another banking professional expressed that "one-size-fits-all" AI marketing may not suit everyone, stating, "Personally, I prefer interacting with a real person."

Killer applications have yet to emerge, while more competitors enter the banking large model arena. A representative from a listed city commercial bank explained that last February, following DeepSeek's launch, their bank also introduced the large model, requiring technical staff to deeply learn and deploy it for development. Regrettably, the bank has not yet developed sufficiently mature or effective applications based on the DeepSeek model. Overall, DeepSeek has not demonstrated significant advantages in banking applications. "The novelty has long worn off." Although the bank still deploys DeepSeek, "expectations have been lowered," with no anticipation of disruptive, killer applications emerging soon. A representative from another regional city commercial bank confirmed that in the first half of last year, upon recognizing DeepSeek's limitations in financial data, they attempted to integrate their existing specialized "small models" with DeepSeek to develop practical applications. So far, results have been limited, requiring further R&D and refinement. Interviews revealed that currently, DeepSeek's primary impact may be stimulating rapid follow-up by competitors, sparking intense competition in finance, healthcare, and other sectors. Since mid-last year, large models including Alibaba's Tongyi Qianwen have been actively exploring collaborations with multiple banks, yielding positive outcomes. Previously, a leading state-owned major bank announced a partnership with Alibaba. "Alibaba has recently collaborated with our bank as well. We have also procured and deployed Alibaba's large model," confirmed a representative from a leading city commercial bank, noting the bank's current strategy of "walking on multiple legs." However, related applications are still under exploration, given banks' higher requirements for data security and system stability.