An AI-Driven Socio-Technical Framework for Performance Management in Teleworking Environments

The shift to teleworking, defined as technology-enabled work arrangements in which employees perform organizational tasks remotely outside traditional office settings, has exposed the limitations of traditional performance management systems, including the lack of direct oversight, micromanagement risks, communication barriers, and employee isolation and well-being. These systems often rely on physical presence or intrusive surveillance rather than outcome-based evaluation. This paper asks how AI-driven performance management can be designed to address the documented challenges of teleworking while safeguarding employee autonomy, fairness, and well-being. The study integrates a comprehensive literature review on AI capabilities with empirical evidence from a sequential mixed-methods study of Canadian public servants, comprising machine learning analysis of over 205,000 tweets, document analysis of federal and provincial teleworking policies, a survey of 176 public servants analyzed using logistic regression, and semi-structured interviews with Government of Canada employees. Grounded in socio-technical theory and the Theory of Planned Behavior, the findings reveal that organizational support, workplace socialization, and attitudes are stronger predictors of teleworking success than digital skills or monitoring, while isolation functions as a measurable risk factor. These empirical patterns are mapped to specific AI capabilities to produce a socio-technical framework organized around three interdependent layers: technological, organizational, and human-centered. The paper contributes an empirically grounded alternative to purely speculative treatments of AI in performance management, offering design requirements derived from what teleworkers actually experience rather than from technological possibilities alone. While the framework is analytically grounded in empirical evidence, behavioral theory, and existing AI capabilities, it has not yet undergone full technical or longitudinal organizational validation. Accordingly, it should be understood as a theoretically and empirically informed design artifact intended to guide future implementation and evaluation efforts. It is worth acknowledging that the study’s key limitations include a Canada-specific public sector sample, modest survey and interview sizes, and the exploratory nature of several proposed AI capabilities; future cross-sectoral, comparative, and longitudinal research is needed to validate and extend the framework.