In the more than a decade since the global financial crisis, banks and regulatory authorities have continued to raise their vigilance towards operational risk. However, they still face challenges in effectively addressing operational risks. According to reports, major global banks have suffered nearly $210 billion in operational risk losses since 2011 . According to Basel II, operational risk is defined as the risk of loss resulting from errors, violations, interruptions, or damage caused by personnel, internal processes, systems, or external events .
Traditional banks primarily identify operational risks through the screening of a large volume of news articles by risk officers. However, this approach heavily relies on the subjective judgment of risk officers and has issues related to subjectivity, high manual costs, and low accuracy. Artificial Intelligence (AI) algorithms can use big data and machine learning techniques  to automate the analysis of operational risk-related data, thus achieving more accurate and efficient risk identification. However, existing AI algorithms require a substantial amount of data, which can be challenging for small and medium-sized enterprises (SME) to obtain.
How to Use ChatGPT
ChatGPT , developed by OpenAI based on the GPT-3.5 architecture, is a chatbot AI assistant. It is trained on a large-scale corpus, giving it the ability to understand and generate natural language. ChatGPT can be used to answer questions, provide advice, and generate meaningful text based on input prompts or context. Therefore, ChatGPT can be used to help SMEs build a banking operational risk recognition platform.
There are mainly two approaches to using ChatGPT to assist SMEs in identifying banking operational risks:
- Generating Training Data: ChatGPT’s text generation capability can be used to create a large volume of text data related to banking operational risks. The specific steps are shown in Figure 1 (A.1-A.4). By providing ChatGPT with some collected text and their operational risks, you can guide ChatGPT to generate a large amount of new text data. Using these generated training data, SMEs can also train their own models for identifying banking operational risks. Figure 2 provides an example of ChatGPT generating news related to banking operational risks.
- Transfer Learning: ChatGPT possesses general language understanding capabilities, allowing it to be transferred from general tasks to specific tasks related to banking operational risks. The workflow is depicted in Figure 1 (B.1-B.2). First, you input the collected or generated relevant text to ChatGPT and inform it about the corresponding banking operational risks to help ChatGPT understand how to identify risks. After that, you can directly use ChatGPT to identify new text inputs. Figure 3 provides an example of ChatGPT directly recognizing banking operational risks.
Impact of ChatGPT
ChatGPT can generate a large amount of meaningful training data, helping SMEs overcome the challenge of limited data. Secondly, ChatGPT’s powerful text understanding capabilities can significantly enhance the accuracy and efficiency of operational risk identification while reducing the impact of human subjectivity. Introducing ChatGPT into banking operational risk recognition can have a range of impacts and effects:
- Engineering Automation and Intelligence: The use of ChatGPT can automate the analysis and identification of a large volume of banking operational risk data, helping banks quickly detect potential operational risks. This automation and intelligence capability can improve the efficiency and accuracy of risk identification while reducing the workload of manual processing.
- Real-time Monitoring and Alerting: ChatGPT can monitor abnormal situations in the banking operations in real-time and issue alerts. By analyzing operational data, it can quickly identify potential risks and notify relevant personnel to take risk control measures promptly, reducing potential losses due to risks.
- Decision Support and Optimization: ChatGPT can provide decision support and optimization advice to bank employees. It can analyze operational data and offer insights and recommendations. This capacity for decision support and optimization can reduce human errors, enhance the quality and efficiency of operations.
However, when using ChatGPT, there are also some potential risks to be aware of, with the most critical being privacy concerns. In the feasible solutions provided earlier, there is a need to input data from your company or collected data into ChatGPT, and the privacy and security of ChatGPT have not been thoroughly evaluated. Therefore, it is essential to pay special attention to privacy issues when using it.
Current State of the Financial Industry
The article  highlights the significant potential of large language models like ChatGPT in the financial sector, explicitly mentioning that AI methods such as ChatGPT can play a crucial role in monitoring and managing banking risks. In Italy’s National AI Strategic Plan launched in November 2021, one of the 11 priority areas is AI in the financial sector. At the Money 20/20 FinTech conference held in Amsterdam in June 2023, executives from global banks and digital financial companies praised generative AI, including ChatGPT .
ChatGPT and other large language models are also rapidly being adopted by FinTech companies, such as Stripe, Klarna, Chime, SESAMm, Tractable, and the AI market in the FinTech sector is expanding at a rate of 28.6%, expected to reach $31.71 billion by 2027 . Online payment provider Stripe has long been using AI to enhance its products and user experience, including helping users manage fraud (as shown in Figure 4) and improving conversion rates. Additionally, an article  presents a customer behavior classification model based on GPT and similar methods for fraud detection, which has been thoroughly validated in Prometeia Associazione’s fraud detection task. The model achieved an accuracy rate of 95.5% in approximately 450,000 synthetic card transactions, outperforming competitors. Furthermore, the model was also tested on an open-source loan default prediction dataset, achieving an accuracy rate of 94.5%.
Generative AI technologies, including large language models, are also piquing the interest of the banking industry. They are actively exploring ChatGPT-like solutions and their potential across various domains. Deloitte’s analysis suggests that by 2026, the use of generative AI can increase the productivity of global bank front-office employees by up to 27%–35% . By introducing ChatGPT, SouthState Bank has seen significant improvements in productivity. For example, tasks that previously took an average of 12 to 15 minutes can now be completed in just a few seconds . A LinkedIn post  explores 15 ways in which ChatGPT can be used in the banking sector (partial examples are given in Table 1). This article also provides two feasible solutions for using ChatGPT to assist in identifying banking operational risks, as shown in Figure 2 and Figure 3.
|Writing a bank marketing copywriting
|Automated Customer Service
|Personalized Financial Advice
Future Challenges and Outlook
It can be foreseen that in the near future, more and more companies will leverage ChatGPT’s text understanding and generation capabilities to assist in identifying potential banking operational risks. We can expect an increasing number of data, models, and platforms to be introduced to help the entire banking industry combat operational risks. ChatGPT’s text understanding abilities can be used to identify potential risks, rapidly construct risk recognition models, and aid banks in establishing corresponding recognition platforms, significantly strengthening their monitoring and management of operational risks.
With ChatGPT’s assistance, we can better mitigate the harm caused by operational risks. Its robust text understanding and generation capabilities will provide businesses and financial institutions with more accurate and efficient means of identifying and managing operational risks, thereby ensuring the robust operation of the financial system.
The work described in this article was supported by InnoHK initiative, The Government of the HKSAR, and Laboratory for AI-Powered Financial Technologies.
(AIFT strives but cannot guarantee the accuracy and reliability of the content, and will not be responsible for any loss or damage caused by any inaccuracy or omission.)