Performance evaluation with algorithms in the loop: How managers evaluate subordinates in AI-supported decision-making contexts

Sammanfattning

Organizations are increasingly adopting AI for decision support, in which algorithms provide recommendations but humans remain responsible for final decisions. While prior research has examined human-AI decision structures, responses to algorithmic advice, and responsibility attribution, little is known about how human decision-makers are evaluated in these settings. To help fill this gap, this thesis aims to address: How do managers evaluate subordinates' performance in AI-supported decision-making contexts? Drawing on fourteen semi-structured interviews with managers from diverse industries and hierarchical levels, the study examines both the performance dimensions managers emphasize and how they combine these dimensions into overall judgments. The findings show that, while outcomes remain central, AI-specific procedures and AI-related capabilities are becoming increasingly salient in performance evaluation. Furthermore, managers tend to adopt an average-like logic to integrate multiple dimensions to form the final judgment. From a theoretical perspective, the thesis extends the management control and outcome bias literatures by showing how AI reshapes the content and weighting of established control types and how outcomes influence not only evaluative judgments but also the criteria considered during evaluation. From a practical standpoint, the study highlights the importance of designing performance evaluation systems that explicitly integrate results, action, personnel, and cultural controls to support effective human-AI collaboration.