A Trust-Based Framework for Explainable AI in Project Management Information Systems in Construction Industry
Pysyvä osoite
Kuvaus
Opinnäytetyö kokotekstinä PDF-muodossa.
The adoption of Artificial intelligence (AI) in Project Management Information Systems (PMIS) has recently become more common. This trend is even more widespread in complex con-texts like construction, where projects are based on huge dataset analysis. However, as AI systems become more developed, some new challenges emerge. Here, trust is a decisive factor that determines whether a project manager can follow AI recommendations. Some concerns in the area of transparency and accountability may hinder the adoption of AI-enabled PMIS. Project managers have to evaluate AI recommendations and then justify their decisions to other stakeholders. In this context, explainability is a must that supports trust. For this study, 10 project managers based in Finland and working in construction industry were interviewed. These interviews were conducted via Zoom and in person, and lasted around 40 to 60 minutes, from February to March 2026. Participants had to meet three criteria: three years of project management experience, familiarity with PMIS, and exposure to AI tools. They came from different sectors of construction industry, including infrastructure, EPCM, commercial, residential, industrial, and consulting. The goal of this study is, first, to understand how project managers gain trust when they use AI-enabled PMIS in their projects and second, to understand what the role of explainability is in building that expected trust. The study uses the logic of the Design Science Research (DSR) framework, which means it is both qualitative and design-oriented. The data were analyzed through thematic analysis in NVivo 15 based on the Stimulus-Organism-Response (SOR) theory. The results show that trust is not just about technical accuracy. Project managers trust AI recommendations when they are transparent enough and aligned with site realities. Moreover, trust is built when system has performed successfully over time. On the other hand, trust reduces when the system lacks explanation and acts as a black box, or when it ignores real site conditions. Trust also suffers when past AI recommendations were faulty, especially when the errors led to financial or schedule consequences. Another key insight is the "site reality gap". Even when an AI recommendation is analytically sound, it can be rejected if it ignores what is actually happening on site. These incidents often involve subcontractor delays, weather conditions, access problems, or stakeholder expectations. Participants emphasized that explainability should support understanding recommendations, assessing uncertainty, and justifying decisions or defending them against clients or stakeholders. Explanations need to be clear and layered, accompanied by key drivers, confidence levels, and visual formats. Based on these findings, this study provides a trust-based explainability framework with five layers: AI capability, explainability design, contextual alignment, accountability and governance, and hu-man–AI collaboration. This framework can be used to develop AI systems that are not just smart, but understandable, defensible, and practical, especially in projects where decisions carry significant weight.