Künstliche Intelligenz in der Entscheidungsfindung: Potentiale und Herausforderungen im Requirements Engineering

Abstract

Requirements Engineering (RE) is a central component of software development, in which decision-making plays a crucial role. Decisions made in the context of RE not only influence the quality of requirements but also determine the success of the entire project. Traditional decision-making approaches in RE are increasingly challenged by the growing complexity of modern software projects. They are often time-consuming, based on subjective judgment, and heavily reliant on individual experience. At the same time, the use of Artificial Intelligence (AI), especially through Large Language Models (LLMs), opens up new potentials for supporting decision-making. However, there is still a lack of solid scientific findings regarding the use of AI in RE-related decision-making, as well as a shortage of empirical studies on its acceptance among practitioners. This thesis aims to address this research gap. The thesis is structured into a theoretical and an empirical part. The theoretical section provides foundational knowledge on AI, RE, and decision-making. It is followed by a detailed analysis of how decision-making occurs within the individual phases of the RE process and to what extent AI can provide support. The focus lies particularly on Natural Language Processing, Machine Learning, and LLMs. The empirical part is based on a qualitative focus group discussion with six experienced RE practitioners. The discussion was guided by a structured interview protocol, transcribed, and analyzed using Mayring’s structuring content analysis. The qualitative analysis was conducted with MAXQDA and structured into four main categories: decision-making in RE, use of AI, challenges, and future outlook. The findings show that decision-making in RE is heavily shaped by collaboration and significantly influenced by the implicit knowledge of stakeholders. AI is not perceived by participants as a replacement, but rather as a “sparring partner” or “coach” that supports reflective decision-making through targeted questioning. AI’s particular strengths lie in identifying inconsistencies and supporting automated analysis and documentation. Key challenges exist in the areas of data protection, transparency, and accountability. The traceability of AI-generated recommendations (“reasoning”) was emphasized as a crucial prerequisite for acceptance. There is skepticism regarding the full automation of creative or strategic decisions. Looking ahead, the participants expect the emergence of new AI systems with enhanced capabilities, while interpersonal aspects are seen as irreplaceable. Overall, the thesis emphasizes that AI-supported systems hold significant potential to improve decision quality in RE, provided they are applied in the appropriate context and used responsibly.

Date of Award	2025
Original language	German (Austria)
Awarding Institution	Johannes Kepler University Linz
Supervisor	Andreas Auinger (Supervisor) & Klaus Lehner (Supervisor)