Large Language Models in Relationship Management: Opportunities and Limitations for Advisory Work in Banking

Topic

Relationship management in banking is highly information-intensive and increasingly shaped by efficiency pressures, customer expectations, and digital transformation. This thesis examines how an internally integrated Large Language Model (LLM) tool influences advisory work within a Swiss bank. Based on semi-structured interviews and qualitative content analysis, the study investigates how advisors use the tool in daily work processes, which advisory activities benefit most from LLM support, and where important limitations remain.

Relevance

Client advisors in banking operate in increasingly complex environments characterized by fragmented information systems, regulatory requirements, and rising expectations for personalized advisory services. At the same time, advisors spend substantial amounts of time on administrative and information-processing tasks rather than direct customer interaction. LLM-based systems offer opportunities to support advisors by simplifying information retrieval, summarizing documents, and improving communication processes. However, their integration also raises questions regarding reliability, practical value, and the future role of human advisors. Understanding these opportunities and limitations is essential for banks seeking to integrate LLM tools into advisory work effectively.

Results

The findings show that the LLM tool was primarily used for information-processing and customer-related tasks, particularly summarization, internal information retrieval, translation, and written customer communication. The strongest perceived value emerged in preparation and follow-up activities surrounding customer interactions. The interviews further revealed that the tool was associated with efficiency gains, improved structure, and more professional communication. At the same time, relationship management remained strongly human-centered. Trust-building, customer understanding, and final decision-making continued to be perceived as core responsibilities of human advisors. The findings additionally suggest that the tool’s current potential has not yet been fully utilized, as adoption depended strongly on role-specific task fit, positive user experiences, and organizational support.

Implications for Practitioners

• Integrate LLM tools mainly for preparation, follow-up, information-processing, and written customer communication tasks rather than positioning them as solutions for fully automating advisory work.

• Develop role-specific training and implementation approaches supported by practical use cases and early positive user experiences.

• Focus on high-quality internal knowledge sources and retrieval-based architectures to improve output reliability.

• Maintain a human-centered focus in areas requiring trust, contextual understanding, and professional judgment.

Methods

The thesis applied a qualitative, exploratory research design based on nine semi-structured interviews with employees from different advisory-related roles within a Swiss bank, including customer book managers, financing specialists, investment specialists, and one LLM implementation analyst. Two use case sections were integrated into the interview guide to reflect different advisory contexts and task characteristics. The interviews were transcribed and analyzed using Mayring’s qualitative content analysis approach supported by MAXQDA. This approach enabled the identification of recurring themes related to communication support, efficiency gains, human-centered advisory work, and organizational adoption patterns.