Artificial Intelligence in Project Management – A Study on the Integration of Large Language Models in the Effort Estimation of Agile Projects

Context

This bachelor's thesis deals with the topic of artificial intelligence in agile project management, an area in which artificial intelligence is becoming increasingly important across all industries. In particular, the role of Large Language Models (LLMs) in supporting effort estimation methods in software companies is examined in more detail. Despite the multitude of advanced methods for estimating project effort, the challenge remains that the accuracy and efficiency of these methods often leave much to be desired. Against this background, the integration of LLMs to optimize these estimation methods appears to be a promising approach.

Goals

The goal of this thesis was to identify possible applications for the integration of Large Language Models (LLM) in the individual process steps of agile effort estimation and to make recommendations for this. In the end, the work should provide an overview of current effort estimation methods and show in which process steps LLMs can be used to improve the accuracy and efficiency of effort estimation. The aim of this work is to show in which process steps and in which effort estimation methods there is potential for the integration of LLMs and which process steps can perhaps even be taken over by LLMs.

Method

In this work, effort estimation methods were first collected from the literature. Criteria were then developed to evaluate how LLMs could support the identified process steps. The creation of a taxonomy served as a methodological foundation. This taxonomy enabled several iterations to be carried out, which provided essential results on the effectiveness of LLM integration in effort estimation.

Results

The iterative process made possible by the taxonomy led to the creation of a matrix. The matrix below has 15 process steps on the X-axis and 19 effort estimation methods on the Y-axis. The cells illustrate the level of support that an LLM can provide for each effort estimation method and process step. Green means that LLMs can take over the process completely, orange means that LLMs can support the process step but cannot take it over completely and red means that LLMs cannot provide any support at all in the process step.

The results show that LLMs can be particularly helpful in collecting data from past projects and in building consensus within the team. Although LLMs could completely take over and automate some process steps, human supervision and involvement remains essential in most cases. It was found that LLMs can generally be used in most process steps and play a supporting role. Especially the preparation of data for further use in effort estimation could be completely taken over by LLMs in many cases.

The integration of LLMs into effort estimation for agile projects presents significant potential and numerous avenues for future research, including the development of advanced LLMs tailored for specific project needs and the exploration of their long-term effects through studies and pilot projects. Additionally, it is crucial to address ethical considerations related to privacy and transparency, which could guide the responsible use of AI in project management.