The Taiwan Banker

Digital transformation in banking over the past decade can be summarized in one word: migration. Moving branch-based services online, and bringing counter services into mobile apps, have been large undertakings. The goal from early online transfers to digital account opening and online lending has always been to change channels, making banking services available anywhere without the constraints of time and location. As generative AI has matured, however, the battleground has quietly shifted. Banks as demand-side players focus on user experience and business outcomes, while technology vendors as enablers understand the underlying architecture and technical constraints. Only by understanding both mindsets at the same time can one see the full picture of AI transformation, helping banks complete the critical evolution from digitalization toward intelligence. To understand the impact of AI, it is necessary to revisit the industry trajectory. Drawing on experience from building an online-only bank in 2016 and leading a digital migration for international card networks at a foreign bank in 2018, the core objective at the time was straightforward: extreme levels of self-service. In other words, the front office revolution has been moving from tool optimization toward intelligent agents. That was the era of tool-centric thinking. Banks focused on moving complex offline processes online. Product managers and designers worked relentlessly on UI and UX, all to eliminate a single click and enable customers to open their account or transfer money as smoothly as possible by themselves. The underlying assumption was that customers knew exactly what they wanted, and the bank’s role was simply to provide the best possible tool. Success was defined by the ease of use and self-service rate. The arrival of generative AI created a qualitative shift. On the service side, AI replaces rigid keyword search with powerful semantic understanding. As one example, Morgan Stanley uses AI to generate instant summaries of research reports for financial advisors. On the marketing side, the focus has moved from simple product sales to lifetime financial planning. Systems can predict life stages through spending patterns, such as calculating education funds for new parents, or proactively offering integrated suggestions before a customer’s financial condition deteriorates. Banks will no longer be cold sales entities, but financial partners who truly understand their customers. Front-office AI is not without risks – most prominently hallucination. Banks can face significant legal liability and reputational damage if their AI systems fabricate wealth management advice to please a customer, invent market data, or misinterpret product terms, so they must take an extremely cautious approach when deploying generative AI directly to consumers. A common transitional model is human-in-the-loop. AI generates a draft recommendation quickly, but the final content must still be verified by an RM or customer service representative before being sent to the customer. The middle office is the bank’s decision brain, covering risk management and compliance. In the past, business units often viewed the middle office as a brake pad. In the AI era, the middle office can evolve into a precision navigation system which tells the front office how to accelerate safely while keeping risks under control. Looking back at the development of personal loan products, risk control largely relied on static labels. Banks tended to target low risk groups such as civil servants and public sector employees, while excluding freelancers with less straightforward income documentation, thereby missing high-quality customers. Today, AI enables dynamic risk control. Models no longer rely only on job titles, but instead analyze unstructured cash flow patterns and transaction behavior. A modern system can determine that a person is a freelance contractor, yet receives stable monthly inflows and has a very low default rate. This improved resolution not only optimizes credit models, but also can become a key technology for financial inclusion. Beyond risk models, one of the most demanding responsibilities of the middle office is money laundering and fraud prevention. Traditional rule-based frameworks, such as alerts for single transfers above a threshold, are easy to evade, generate large volumes of false positives, and are becoming less effective. AI models are good at handling complex relationship networks and anomaly detection. They can compare the characteristics of a payee account with the customer’s past behavior patterns within milliseconds from the moment the transfer is initiated, judge risk in real time, and block the transaction. By analyzing fund flow graphs, they can improve the identification of hidden mule accounts. The middle office carries many highly knowledge-intensive, yet extremely cumbersome tasks. In corporate credit and compliance reviews, for example, reviewers must read hundreds of pages of reports. AI with advanced intelligent document processing and natural language processing capabilities can rapidly digest massive unstructured documents, automatically extract key data, and generate an initial report. This does not imply that middle office professionals will lose their jobs; on the contrary, when AI takes on 80% of the organization work, experts can finally focus on the 20% of decisions that are most complex and require experience-based judgment and commercial negotiation. The back office may operate out of sight, meanwhile, but it determines whether a bank survives. The core accounting systems that support Taiwan’s financial system still largely run in COBOL environments. Maintenance of COBOL systems is becoming an increasingly concentrated and specialized domain. Banks are working to systematically transfer senior experts’ tacit knowledge, while technology vendors are reallocating the skill focus of younger talent, responding to changes in the broader ecosystem. Against the backdrop of workforce restructuring, optimization of resource allocation to ensure stable continuity of core systems is a major issue for both sides in the modernization journey. Many across the industry are exploring whether AI can serve as a universal engineer, assisting with maintenance of legacy systems or language conversion, such as COBOL to Java. The question is whether AI can truly understand decades of layered and obscure logic: in practice, it can indeed accelerate translation and annotation, but in an environment where core systems cannot tolerate any error, AI-generated code must undergo extremely rigorous verification. AI brings hope to core modernization, but it is a surgical instrument that requires skilled craftsmen to operate with precision, not a one-click upgrade. Ultimately, the success or failure of AI transformation often depends less on the technology itself, and more on how management responds to the organizational and governance challenges that follow. AI feeds on data, yet many banks have data scattered across isolated systems in inconsistent formats. The winners will be the institutions willing to break through the underlying data architecture and build unified data platforms. This is a foundational infrastructure project. The changes wrought by AI inevitably trigger internal anxiety, which requires leadership to redesign workflows and build a culture of human-machine collaboration. Consider a protection framework using multimodal AI. By combining facial recognition, core banking system data, and AML systems, the system can do more than simply display a error code when a branch counter detects abnormal customer expressions, or when facial features trigger a spoofing alert. Through generative AI integrated into the teller’s interface, the system can recommend guidance scripts in real time, such as alerting staff of risk, and advising them to politely ask about the purpose of the remittance, or to attempt to delay while notifying relevant teams. In this setup, AI handles the computationally-intense judgment, while staff handle the communications requiring emotional intelligence. Future bankers will require less rote memorization, and more judgment and interpersonal connection. Serious ethical issues arise when AI algorithms influence asset allocation decisions. Financial professionals must carefully assess the appropriate boundaries of AI use, especially when supporting high net worth clients. It should be positioned as a copilot who provides data insights and scenario simulations, not as the driver holding the final steering wheel. Only when AI is grounded in ethics and explainability can it earn long-term trust. This decade shows that technology is a means, not the end. Future competition among banks will be determined by whose back office is more resilient, whose middle office is more precise, and whose front office understands customers better. A shift is underway in the role of AI in finance from an optional add-on toward becoming a baseline requirement. The author is a Senior Sales Manager for Digital Finance at NEC.