The waves created by the rise of artificial intelligence (AI) in China are undeniable, with a prominent surge observed in sectors like finance. A recent assertion by UBS paints a vivid picture, indicating that due to the data-heavy and labor-intensive nature of the financial industry along with numerous linguistic tasks involved, it is anticipated that generative AI (GenAI) will reshape this sector more profoundly than others.

In fact, a steady increase in investments towards information technology has been evident within the banking sector in China. For instance, in 2021 alone, the total investments from 18 listed banks reached a staggering 158.4 billion yuan. However, as we look towards the future in 2023, the total investments across 17 banks have jumped to 184.7 billion yuan. Notably, around 20 financial institutions, including major players like ICBC and Agricultural Bank, have already begun to disclose their advancements in large-model research, practical applications, and business empowerment in their annual reports for 2023.

Nevertheless, the adoption of AI is not without its challenges. Issues such as AI model hallucinations, ethical biases, and data privacy concerns have raised eyebrows. Yet industry experts emphasize the importance of embracing these innovative technologies. Stagnation born from fear of potential challenges could lead to a detrimental lag in the competition.

Zack Kass, former Global Market Application Leader for OpenAI and now an AI and business strategy expert, shared in an exclusive interview that he sees generative AI enhancing efficiency across five critical areas in the financial industry: retail banking, wealth management, underwriting and pricing of insurance products, sales and marketing, and investment banking. These sectors, characterized by high personnel demands and the necessity for specialized expertise, highlight how AI could significantly uplift ROI in some historically inefficient banking operations.

Delving deeper into these five areas, retail banking stands out prominently. It remains a fiercely contested domain for banks; however, it often entails high costs and low returns on investment. Maximizing the ROI of low-margin businesses through AI could represent a significant breakthrough. The optimization that AI brings can improve the quality of retail products while simultaneously lowering costs and boosting efficiency. Retail banks also grapple with challenges such as high employee turnover rates and varying work quality—factors that AI has the potential to mitigate.

Wealth management emerges as the second key area where the application of AI is expected to make substantive changes. Exceptional talent in wealth management is scarce globally; hence, leveraging AI can empower wealth managers in two critical ways: expanding their service capabilities and standardizing those services to uplift the overall professional bar across teams. Moreover, AI can help identify which wealth managers might struggle to succeed, allowing for a more effective allocation of resources.

The third area where AI cannot be overlooked is underwriting and insurance product pricing. By employing more precise data analytics, AI addresses several complex challenges faced by the insurance industry, particularly in risk pricing and claims prediction, thus enhancing effectiveness across the board. This technology holds promise in refining pricing methodologies and making lower-value insurance products viable. Besides, platforms like ChatGPT can bolster customer experiences, thereby amplifying market penetration of insurance products.

When it comes to sales, marketing, and customer acquisition, AI serves as a tool to refine sales strategies, subsequently driving up conversion rates and market share.

Investment banking, the fifth category, stands to benefit immensely from AI as well. AI can handle vast amounts of complicated financial data analysis, including historical transaction records and corporate financial statements. Advanced machine learning algorithms will unveil hidden patterns, lending banks constructive capabilities to predict market trends more accurately. Furthermore, natural language processing (NLP) technology can efficiently scan and analyze news articles, financial reports, sector analyses, and social media content for valuable insights. Financial modelling, which often consumes considerable human effort, can be expedited dramatically with AI's capabilities to construct and enhance financial models by performing extensive calculations almost instantaneously.

However, as institutions dive into AI explorations, multiple obstacles still linger. Financial institutions are actively probing into the manifold applications of AI, but the practical implementation rate remains limited compared to international standards.

According to analyst Cao Haifeng from UBS Securities, a significant revelation highlights that financial institutions are still mainly focused on employee empowerment applications, such as knowledge assistance, customer service aids, smart investments and research, and insurance agent tools. In this ecosystem, human oversight remains pivotal. Larger financial firms are systematically developing large language models (LLMs) that integrate front, middle, and back-office functions while employing various modeling approaches, including in-house and collaborative development. Conversely, smaller firms are cautious due to cost constraints, often opting for straightforward scenarios.

Cao elucidates that suboptimal effectiveness of domestic models, high threshold for investment, and ambiguous ROI are critical factors inhibiting the widespread deployment of generative AI in finance.

Moreover, personnel from a leading joint-stock commercial bank shared ongoing concerns over the continual influx of orders for large model services, despite the high costs associated with these products. While the potential for large models is acknowledged, many banks report insufficient budgets to keep up with the accelerating pace of industry evolution.

Overall, the financial sector appears to grapple with a paradoxical attitude towards large models. Training these complex systems is resource-intensive and costly, leaving most institutions overwhelmed. Reports indicate that only a handful of banks are advancing towards independent large-model applications. The majority remain engaged with third-party model firms, acquiring cloud-based models as needed. This hesitation is fueled by the myriad challenges associated with the practical deployment of large models across numerous dimensions, including hallucinations, data privacy concerns, and weak interpretability. At a business level, substantial cost investments do not guarantee positive ROI, creating further doubts.

High Xu Lei of China Merchants Bank discusses the growing pains of AI, emphasizing a triad of significant challenges: first, the phenomenon of model hallucinations can lead to misleading conclusions and potentially severe repercussions; secondly, ingrained value and ethical biases within AI models could inadvertently promote harmful content; and finally, the risk of customers’ privacy being compromised, alongside the possibility that efficiency may gravitate towards a 'winner-takes-all' effect.

High notes the necessity for banks to unravel the conundrum of achieving "more, better, and cheaper" within the paradigm of plentiful clients, supreme service experiences, and minimized costs. This balance, he asserts, is fundamentally challenging; managing a large clientele often incurs high-risk elements, translating to elevated expenses, while reducing costs can detrimentally affect customer experience.

In recent developments, several banks have initiated large model-related infrastructure projects. China Merchants Bank, for instance, embarked on a 'Smart Computing Platform' earlier this year, specifically targeting the development of financial vertical-large models. Some early outcomes have already emerged, such as automated marketing copy generation and significantly enhanced efficiency through AI-enabled customer service platforms capable of addressing customer inquiries in seconds rather than minutes. They have also rolled out a robust risk control platform that integrates numerous neural network algorithms.

Focusing beyond the immediate effects on the financial industry, one cannot ignore the symbiotic relationship emerging with AI technology suppliers. Financial IT firms stand to gain the most from the integration of AI, utilizing their accumulated data assets and sector insight to drive innovation.

Cao speculates that as AI applications scale, financial IT companies are poised to become major beneficiaries, particularly given their adeptness at aligning large language models with the requirements of financial institutions. Forecasts suggest that by 2030, generative AI could substantially influence the financial IT industry, with projections indicating revenues could reach a staggering 69 billion yuan, constituting 24% of total industry revenue, benefiting the banking, insurance, and financial IT realms equally.

Additionally, UBS expresses optimism towards specific firms such as Yuxin Technology, recognized as a banking IT leader, along with Hengsheng Electronics as a primary player in the securities IT landscape. While some financial information providers demonstrate progress with generative AI faster than anticipated, they must navigate potential challenges affecting core business trajectories.

In essence, Yuxin Technology, which offers an all-in-one banking product solution, has introduced several applications, including low-code application development platforms, data security classification assistants, and intelligent customer service tools. Hengsheng Electronics has been proactive in providing solutions tailored for securities language models, notably contributing to the latest FundGPT model developed in collaboration with Industrial Bank of China. Recently, the company's announcement of fully integrating large models into its offerings underlines the advancing frontier of AI applications.

In a future dominated by generative AI, significant transformations await the financial domain. Currently, UBS has identified an impressive 31 and 23 applications within the brokerage and insurance sectors, envisioning substantial improvements in cost reduction, efficiency enhancement, and client loyalty rates, with a notable prediction that labor cost medians could drop by 20% by 2030.

Specifically, regarding the brokerage sector, anticipated generative AI applications may enhance return on equity (ROE) by 1.4% by 2030, concurrently propelling net profits by 15% and catalyzing a forecasted valuation growth of 19% for that year. On the insurance side, it is predicted that generative AI applications could contribute to a 17% increase in new business value (NBV) by 2030, heralding an 12% growth in valuations.