A survey by consulting firm McKinsey & Company found that a majority of Taiwanese banks have begun applying or testing Generative AI (GenAI).
The financial industry estimates large-scale applications of AI to materialize within one to three years. Experts caution that using AI involves more than just choosing the best Large Language Model (LLM) and the related cost for system integration should also be considered.
GenAI in the financial industry
McKinsey's first survey on the application of GenAI in Taiwan's banking sector included senior representatives from 15 banks, whose combined net income accounts for about 70% of the industry's total.
The results showed that 75% of banks have started applying GenAI through various pilot projects, but fewer than 10% have achieved large-scale application.
Common application scenarios include operations and services, marketing, IT development, compliance auditing, risk management, and talent management. The most prevalent use is in operations and services, such as internal knowledge virtual assistants, providing employees with relevant business knowledge, and basic Q&A applications.
Charles Tan, a McKinsey partner, pointed out that the banking industry requires many IT personnel. Older programming languages like Cobol are increasingly lacking maintainers. More GenAI tools now support coding tasks, accelerating processes that would have taken five to ten years to convert programming languages previously.
LLMs are a whole package
McKinsey senior partner Violet Chung highlighted that GenAI is not just about LLMs but an integration of many capabilities. She explained that while some firms focus solely on finding the best solution, they overlook other critical issues. She likened LLMs to the human heart, which needs a network of veins and systems to function properly.
Chung emphasized that purchasing solutions is just the first step. Regardless of the provider, the basic package involves organizational transformation, localization, and talent training. There is a significant gap between acquiring services and achieving effective application.
Tan, on the other hand, observed that many commercial software codes are unmodifiable, so firms capable of fine-tuning models will use ready-made LLMs and adjust them as needed. Open-source models, which offer more customization flexibility, are increasingly popular worldwide.
Regarding the question of whether banks need to purchase additional AI servers or AI PCs, Tan explained that for every dollar spent on specific GenAI software and hardware resources, another five dollars might be needed for related system integration. That expenditure would encompass data, IT architecture improvements, process changes, and employee training.
LLMs alone are not the complete solution.
The race for AI integration
Victor Kuan, a senior advisor at McKinsey Asia, noted that GenAI has become a "must-win battle" for many CEOs, with concerns that competitors will launch new services first and attract their customers. However, he stressed that AI budget planning requires a strategic framework from headquarters or financial holding companies, rather than decentralized procurement by various departments or treating it as a general technology project.
The survey also found that in addition to internal factors, external challenges such as regulations, technology, and talent pose significant obstacles for banks.
In 2023, Taiwan's Financial Supervisory Commission (FSC) announced a draft principle for using AI in the financial industry. After a public comment process, the formal guidelines are expected to be issued in June 2024.
Although these guidelines will be advisory and not legally binding, some firms are concerned that the FSC may influence through administrative inspections. The McKinsey team hopes for faster responses from regulatory bodies and clearer scope definitions.
