Subcontractor evaluations are a critical component of post-project reviews in the construction sector, significantly influencing future procurement decisions, subcontractor selection, relationship management, and risk mitigation strategies. Despite their strategic value, existing subcontractor evaluation practices often suffer from several persistent shortcomings: high subjectivity, lack of standardized performance metrics, limited documentation, exclusion of subcontractor perspectives, and minimal use of structured or historical evaluation data. In response to these challenges, this thesis examines whether and how integrating ChatGPT 4.0 with a Multi-Criteria Decision Analysis (MCDA) framework can enhance the consistency, objectivity, and strategic value of subcontractor evaluations within HOCHTIEF Nederland.
Building on a comprehensive literature review and exploratory interviews with key procurement and project staff, the study identifies gaps in traditional evaluation processes, particularly the overreliance on static rating templates and the absence of dual-perspective analysis. To address this, a novel AI-enhanced evaluation framework is developed using a design science methodology. The framework enables structured performance assessments based on both internal HOCHTIEF feedback and subcontractor self-evaluations. ChatGPT 4.0, embedded in HOCHTIEF’s internal AI assistant “NextChat” supports this process by interpreting qualitative data, prompting for clarification when needed, generating justifications for ratings, and summarizing insights into actionable reports.
The framework was implemented in a pilot case study involving a subcontractor working on a real HOCHTIEF data centre project. Evaluation inputs from both parties were processed using the AI assistant and benchmarked against HOCHTIEF’s existing manual evaluation methods.
Validation of the framework was multi-faceted. Process validation demonstrated that AI-generated reports aligned closely with manual evaluations, with a Mean Absolute Percentage Deviation (MAPD) of less than 10%, indicating high accuracy. Stakeholder validation was conducted through structured surveys with HOCHTIEF personnel, assessing insightfulness, transparency, clarity of follow-up queries, added value, and perceived limitations. The results were consistently positive.
While the results indicate strong potential for improving subcontractor evaluations through AI integration, the study also highlights critical limitations and risks. These include the need for high-quality, context-rich input data, the need for human oversight and verification of the results, data availability challenges, and the necessity of embedding the tool within existing procurement databases to ensure organizational consistency. Furthermore, successful implementation depends on training and change management strategies, particularly in organizations with limited digital procurement maturity.
The study contributes theoretically by empirically validating the integration of AI and MCDA in construction procurement and practically by providing a scalable tool that enhances the structure, transparency, and usability of post-project subcontractor evaluations. It is supported by established adoption frameworks, including the Technology–Organization–Environment (TOE) model and the Unified Theory of Acceptance and Use of Technology (UTAUT). Overall, this research offers a data-informed, dual-perspective, and AI-supported approach to subcontractor assessment, positioning it as a robust enhancement to current construction management practices.