Navigating the maze: the effects of algorithmic management on employee performance

The rapid advancement and widespread application of digital technologies across various industries have reshaped the landscape of work and employment, transforming organizational practices and employment patterns. These changes are pronounced in HRM, where algorithmic management—algorithms that automate or support managerial decision-making—has become a major driver of change by assigning tasks, evaluating performance, and streamlining operations. While algorithmic management offers efficiency and accuracy, its impact on individual employees remains debated. Prior work has emphasized autonomy, rights, and psychological consequences, but less is known about impacts on creative performance (generating novel, valuable ideas) and adaptive performance (adjusting effectively to change). Both are critical in dynamic environments for innovation and resilience. To address this gap, the study builds a moderated mediation model grounded in AMO theory, positing improvisation capability as a key mediator and algorithmic dependence as a boundary condition. The study aims to clarify how algorithmic management influences employees’ creative and adaptive performance, to unpack the underlying mechanisms, and to identify conditions under which effects are amplified or weakened.

The study draws on AMO theory, which posits that performance is shaped by ability, motivation, and opportunity. Prior literature characterizes algorithmic management as involving monitoring, control, task assignment, and performance assessment through rules and data-driven systems, with potential negative attributes such as excessive monitoring, dehumanization, and opacity. These features can reduce autonomy and increase behavioral rigidity, theoretically undermining improvisation capability—the spontaneous, creative reconfiguration of resources to address unforeseen challenges. The authors hypothesize: H1, algorithmic management negatively relates to improvisation capability, due to rigid, historically patterned work guidance, reduced autonomy for intuitive judgment, and fixed task assignment. They further argue, based on research linking improvisation to creativity and adaptability, that improvisation capability should positively relate to creative performance (H2) and adaptive performance (H3). Consequently, improvisation capability should mediate the effects of algorithmic management on creative and adaptive performance (H4, H5). Finally, drawing on the motivational component of AMO, algorithmic dependence (the extent employees rely on algorithmic technology) is proposed to strengthen the negative effect of algorithmic management on improvisation (H6) and thereby amplify the indirect negative effects on creative and adaptive performance (H7, H8).

Design and setting: A multi-wave survey was conducted in an information technology service firm in northern China that uses a well-developed algorithmic management system. The research team coordinated with the CEO and HR manager to recruit participants and administer surveys. Sample and procedure: 453 employees were invited; 327 valid responses were obtained (response rate 72.19%). To reduce common method bias, data were collected in three waves one month apart: Time 1 collected algorithmic management, algorithmic dependence, and controls; Time 2 collected improvisation capability; Time 3 collected creative and adaptive performance. Confidentiality and anonymity were assured; participants completed surveys independently in a controlled setting and received a small incentive (10 yuan). Participants: 21.10% male; mean age 26.99 years (SD = 3.33). Job tenure: 54.73% had ≥3 years with the organization. Education: 85.93% bachelor’s degree; 9.17% master’s or higher. Measures: All scales used a 5-point Likert scale (1 = strongly disagree to 5 = strongly agree) and were translated via back-translation.

• Algorithmic management: 20-item scale (Parent-Rocheleau et al., 2023) covering algorithmic monitoring, goal-setting, scheduling, performance management, compensation (α = 0.889).
• Improvisation capability: 7-item scale (Vera and Crossan, 2005), adapted to individual level (α = 0.930).
• Creative performance: 7-item scale adapted from Gong et al. (2009) (α = 0.950).
• Adaptive performance: 14-item scale adapted from Han and Williams (2008), individual level (α = 0.965).
• Algorithmic dependence: 7-item scale adapted from Shu et al. (2011), substituting “algorithm systems” for “computer technology” (α = 0.937). Controls: Age, gender (1 = male, 2 = female), education (1–5), and job tenure (categorical: within 1 year to >10 years). Analysis: Preliminary correlations were examined; confirmatory factor analysis assessed discriminant validity using item parcels for model parsimony. Hypotheses were tested via structural equation modeling in Mplus 8.3. Software used included SPSS 27.0 and Mplus 8.3.

• Measurement validity: Five-factor CFA model (algorithmic management, improvisation capability, creative performance, adaptive performance, algorithmic dependence) showed excellent fit: χ² = 1208.058, df = 655, CFI = 0.949, TLI = 0.945, RMSEA = 0.051, SRMR = 0.036, outperforming alternative models.
• H1 supported: Algorithmic management negatively related to improvisation capability (β = -0.491, SE = 0.057, p < 0.001).
• H2 supported: Improvisation capability positively related to creative performance (β = 0.381, SE = 0.068, p < 0.001).
• H3 supported: Improvisation capability positively related to adaptive performance (β = 0.410, SE = 0.051, p < 0.001).
• H4 supported (mediation on creativity): Indirect effect of algorithmic management on creative performance via improvisation capability = -0.187 (SE = 0.043, 95% CI [-0.270, -0.104]).
• H5 supported (mediation on adaptability): Indirect effect on adaptive performance via improvisation capability = -0.201 (SE = 0.036, 95% CI [-0.272, -0.131]).
• H6 supported (moderation): Algorithmic dependence strengthened the negative relationship between algorithmic management and improvisation capability (interaction β = -0.245, SE = 0.098, p < 0.05).
• H7 supported (moderated mediation on creativity): Indirect effect significant at high dependence (+1 SD: -0.280, SE = 0.069, 95% CI [-0.415, -0.145]) but not at low dependence (-1 SD: -0.094, SE = 0.049, 95% CI [-0.190, 0.003]); difference = -0.187 (SE = 0.084, 95% CI [-0.351, -0.022]).
• H8 supported (moderated mediation on adaptability): Indirect effect significant at high dependence (+1 SD: -0.302, SE = 0.058, 95% CI [-0.415, -0.189]) but not at low dependence (-1 SD: -0.101, SE = 0.052, 95% CI [-0.203, 0.001]); difference = -0.201 (SE = 0.083, 95% CI [-0.363, -0.039]).

The findings clarify how algorithmic management shapes employee outcomes critical for innovation and organizational resilience. In AMO terms, algorithmic management (opportunity) constrains employees’ improvisation capability (ability), which in turn suppresses creative and adaptive performance. Moreover, employees’ algorithmic dependence (motivation) exacerbates these effects by increasing acceptance of algorithmic control, reliance on system-provided information, and fixed response patterns, thereby further limiting improvisation. Theoretically, the study extends algorithmic management research beyond autonomy and well-being to performance outcomes, opens the “black box” by identifying improvisation capability as a mechanism, and introduces algorithmic dependence as a key boundary condition. Practically, organizations should implement algorithms in ways that preserve discretion and foster improvisation (e.g., training, recognition systems, mixed human-algorithm decision practices) and monitor and manage employees’ overreliance on algorithmic systems to protect creativity and adaptability.

Grounded in AMO theory, the study demonstrates that algorithmic management can diminish employees’ improvisation capability, leading to lower creative and adaptive performance, with stronger negative effects among employees who are highly dependent on algorithms. The work advances the literature by identifying improvisation capability as a mediating mechanism and algorithmic dependence as a moderator. Future research should test generalizability across sectors and cultures, use multi-source and longitudinal designs, and examine additional boundary conditions (e.g., algorithm transparency) and contextual or individual factors that may mitigate or exacerbate these effects.

• Single-firm, single-country sample (IT services firm in China) limits generalizability; cultural and industry contexts may shape responses to algorithmic management.
• Self-reported measures, despite temporal separation, can introduce self-description bias; future studies should include supervisor or peer ratings for performance outcomes.
• Cross-sectional/time-lagged survey design limits causal inference; longitudinal or experimental designs are recommended.
• Focus on algorithmic dependence as the sole moderator; other boundary conditions (e.g., perceived algorithm transparency, environmental and individual differences) warrant examination.