Leveraging AI for Team Building

Abstract

As teams are increasingly used in project-based work, education, and organizational settings, forming suitable teams remains a difficult decision-making problem. Team effectiveness depends on more than individual competence, since many important qualities of teamwork develop through interaction. At the same time, artificial intelligence and people analytics offer methods for processing profile information and comparing possible assignments more systematically. This thesis examines how AI can support team building without treating team effectiveness as fully predictable from pre-formation data.

The thesis follows a design science research approach combining literature-based analysis, computational modeling, artifact development, and technical evaluation. Based on research on team effectiveness, AI-supported personnel decision-making, and computational team formation, the study develops a bounded decision-support artifact for short-lived project teams formed under limited prior knowledge of participants. The artifact consists of a formal allocation method and a Python prototype that allocates candidates to AI-generated and manually reviewed project briefs. The method combines profile-derived candidate representation with weighted candidate-project scoring, round-based marginal-contribution construction, allocation-level fairness scoring, and local improvement. The prototype is evaluated against baseline methods and ablation variants under controlled input conditions.

The evaluation shows that the prototype produced feasible allocations in all tested runs. The marginal contribution construction produced substantially stronger initial allocations than direct-fit and literature-inspired k-rounds construction methods. However, when the same local improvement procedure was applied to the k-rounds allocation, the final results were comparable to those of the full thesis algorithm. This indicates that both the initial construction logic and the refinement stage contribute to allocation quality, while also showing that local improvement can compensate for a weaker initial allocation under the tested conditions. The outputs were also inspectable, since the system reported team scores, skill coverage, missing requirements, baseline comparisons, and local improvement effects. These results do not prove that the recommended teams would perform better in real collaboration. Instead, they show that AI-supported team formation can be implemented as transparent decision support for structuring available information and comparing allocation alternatives.