AI cases in banking – AI radar of the Business Engineering Institute St.Gallen

The financial industry is beginning to undergo a profound transformation through the use of AI. In particular, the use of generative AI opens up the possibility of creating new content such as texts, images, videos or music. In autumn 2024, we started building an AI radar that systematically analyzes use cases of banks in German-speaking countries. This blog post offers an overview of use cases and maturity levels in the banking sector and provides exciting insights into the status quo.

The journey of AI in banking began back in the 1980s, e.g. with the use of AI programs for tax and financial advice. For example, the customer provides information on their sources of income, level of expenditure, assets, liabilities, insurance cover, pension benefits (pension and insurance), risk tolerance and life goals in a detailed questionnaire and a discussion with a financial planner. Goals can include, for example, the standard of living now and in retirement, but also the age of retirement, adequate insurance cover and financing children’s studies. The system determines the archivable level of these goals and develops a customized strategy for achieving them.[1]

The development of algorithms for risk modeling and market analysis (based on labeled data and machine learning (ML)) in the 1990s was followed by the use of neural networks and deep learning (DL) techniques for fraud detection and automated services. In the 2010s, the use of AI was extended to chatbots, robo advisors and personalized financial advice. The use of ML and natural language processing (NLP) improved customer interaction and efficiency. Since the 2020s, the financial industry has increasingly relied on AI for automation, data analysis and compliance. Deep learning and neural networks can be used to analyze complex data volumes and optimize processes.

With the introduction of Large Language Models (LLMs), the framework parameters have changed again and numerous new use cases based on unstructured data have become possible:[2]

  • Intelligent customer interaction: Chatbots and voicebots that understand and process complex inquiries in natural language are revolutionizing customer service.
  • Fraud detection: Real-time transaction analysis enables more precise and faster identification of suspicious activity.
  • Personalized advice: By analyzing customer preferences and market trends, banks offer tailor-made financial solutions.
  • Compliance automation: regulatory reports can be created and analyzed efficiently.
  • Innovative financial products: New, AI-based products open up previously untapped market segments.

AI radar: methodology and findings

The AI Radar is based on a systematic analysis of banks in the DACH region. The banks supervised by the ECB in Germany (65) and Austria (75) were analyzed on the basis of publicly available sources in the context of their use of AI. The same was done for the 15 largest Swiss banks. A total of 61 specific AI use cases were identified. These were then categorized in the Bank Model developed at the Business Engineering Institute St. Gallen according to the processes involved (management processes, front office, middle office, back office and support processes).[3]

Figure 1: Structuring of the AI use cases based on the Bank Model

Each use case found by a bank is further classified in detail, including by maturity level, technologies used (e.g. ML, NLP, DL), development approach (internal, external or hybrid) and other factors.

Due to the exclusive use of publicly available information to date, the first version of the AI radar should be viewed with caution. It is noticeable, for example, that many use cases appear in the front and middle office areas in particular – but significantly fewer in the other areas. This is certainly due to the fact that there is naturally much more communication in the first areas. Either customers are informed about the use of AI-based chatbots etc,. or products that have been developed with the help of AI are presented proactively. From various discussions with bank representatives, it became clear that many banks are already introducing or have already introduced AI, particularly in their back office, for example to make it as easy as possible for employees to query instructions.

Figure 2: AI radar (status 9/2024)

Some exciting use cases appear in the AI radar. In the area of control processes, Commerzbank uses GenAI (ChatGPT) to derive ECB interest rate forecasts.[4] In the front office, there are mainly use cases for the deployment of chatbots – for example, Baloise Bank uses a bot in e-banking that proactively contacts users with a mortgage and offers an extension option.[5] In the Middle Office, Deka is researching the use of BERT to analyze investment decisions. The aim is to determine the sentiment in mandatory announcements from companies. Specifically, readability, such as sentence length, the proportion of technical terms and the ratio of numbers to words are analyzed with the help of NLP.[6]

Looking at the individual process levels, the following picture emerges: In the front office, the focus is on generic chatbots and specific questions or target groups or technologies with over 50% LLM use. Examples include generic chatbots and bots in the context of mortgage offers and renewals, Twint app[7] , chatbot/search service for third parties (e.g., also journalists, applicants), avatars and even ordering account balances via voicebot.

In the back office, the focus of AI cases is on credit checks, ESG reporting and optimizing the customer approach with less than 10% LLM input. In the middle office, we see a focus on market forecasts, report preparation and generic or specific risk management. Here, LLMs are used in more than 10% of use cases. Use cases include the forecasting of market developments (share prices, commodities & currencies, indices, liquidity), the preparation of fundamental financial analyses, programming, corporate governance (risk identification) as well as the detection of insider trading and the optimization of processes.

In terms of support processes, we see a variety of application constellations, with more than 20% of cases using LLMs. Use cases can be found in the areas of marketing, expense management & internal avatar, cloud migration, transformation, cyber attacks/fraud processes as well as in the context of improving IT & data-driven business.

The AI radar is being continuously developed. Based on the radar’s previous focus on publicly accessible information, the next step will be to survey the partners of the Competence Center Future Financial Services (15 banks and providers from the DACH region) in detail about their use of AI. In addition, the focus will be expanded to include banks from the eurozone in order to capture further innovative use cases.[8] Another aspect that will be integrated is the evaluation of the supply side, such as IT providers in the financial industry, in order to take innovative solutions into account, possibly even before they are implemented.

Many banks that lack experience in dealing with “more sophisticated” AI start with the back office and support processes and only approach the front office once there is sufficient expertise and cultural readiness. This also prevents potential risks arising from the use of AI from being experienced by the customer.

Conclusion

The AI Radar 2024 provides an overview of which bank has implemented which AI use cases for which processes. Further information such as the type of AI, provider and adaptation effort is also recorded – where communicated.

The potential of AI from intelligent chatbots to market analysis is enormous. However, in order to be successful in the long term, banks need to invest not only in technology, but also in their employees and their skills and mindset, as well as in collaboration with other companies.


[1] Kindle et al (1989). PFPS – Personal Financial Planning System. IAAI-89 Proceedings. P. 38-44. https://cdn.aaai.org/IAAI/1989/IAAI89-007.pdf

[2] For a detailed analysis of LLMs, see the three-part Large Language Models: Emergence-Use-Further Development | cc-bei.news.

[3] Zerndt (2020). Bankmodell. Gabler Banklexikon. https://www.gabler-banklexikon.de/definition/bankmodell-70673

[4] Wagner (2024). How ChatGPT helps with ECB forecasts. Commerzbank. https://www.commerzbank.de/konzern/newsroom/publikationen/240726wifchatgpdezbprognosen.html

[5] Baloise (undated). https://www.baloise.ch/de/privatkunden/konten-karten-finanzierung/services/e-banking-baloise/chatbot/bedingungen-und-datenschutzhinweise.html

[6] Deka (n.d.). Added value through artificial intelligence. https://www.deka.de/deka-gruppe/media–research/was-uns-bewegt/mehrwert-durch-kuenstliche-intelligenz

[7] LUKB’s chatbot “Lou” is currently answering questions about TWINT, including the TWINT download option. https://www.zugerkb.ch/zugerkblog/posts/kennen-sie-mona

[8] Evident (2024). The Evident AI Index. https://evidentinsights.com/ai-index/

Stefanie Auge-Dickhut