Algorithmic Management Dilemma: The Double-Edged Sword Effect on Employee Well-Being and Task Performance

Shili Luo

School of Economics and Management，Sichuan Agricultural University, Mianyang, 621000,

Sichuan Province, China.
v455921738@163.com

Abstract
An algorithmic management (AM) has emerged in the rapidly changing arena of digital
workplaces, representing a transformative — and controversial — concept. Using a survey with
312 respondents from a variety of industries where work is allocated, monitored and evaluated
using algorithmic systems, this study examines the double edge sword of AM impact on
employee well-being and task performance. The study shows that while AM significantly
increases task performance (measured by higher efficiency and consistency; better real time
feedback), it significantly but negatively affects employee well being (higher psychological
stress, perceived surveillance and autonomy loss). The mediating role of perceived autonomy in
the relationship between AM and well-being is further confirmed by the mediation analysis and
it reveals its crucial role in human systems interaction. This is a paradox: algorithmic tools that
are built for performance optimization can cause emotional exhaustion and alienation when
implemented without human-centered defenses. This study enlarges the ongoing discussion of
digital labor, AI machine work systems and workers’ resiliency and provides practical insights
for ethical design and regulation of AM technology. The need is for an ‘and’: the harmony of
automation and empathy to guarantee that the primal essence of human experience in work is not
digested by the efficiency of technology.

Keywords: Algorithmic Management, Employee Well-Being, Task Performance, Autonomy,
Digital Labor, Surveillance, AI in HR, Ethical Design, Workplace Automation, Technostress

1. Introduction
   Rather than an exception, algorithmic management (AM) has become a force transforming how
   work is organized, supervised and evaluated in the digitizing landscape of work. The term
   Algorithmic Management (AM) was coined by Lee at al. (2015, p. 645) as the use of algorithms
   and data driven technologies to coordinate and manage human labor (see also Keeling, 2019) and
   has been adopted widely across industries, from the gig economy platforms such as Uber and
   Deliveroo (Newlands, 2021), to formal corporate environments, for example in logistics, retail
   and customer support (Kellogg et al., 2020). Algorithmic management essentially substitutes or
   augments these functions: task allocation, performance evaluation and behavioral oversight by
   automated, algorithmic decision making systems that operate in real time (Mateescu & Nguyen,
   2019). While it promises increased efficiency, scalability and objectivity, it also presents unique
   ethical and psychological challenges associated with employee well being and autonomy.
   But the growing literature on algorithmic management has begun to take stock of the ways in
   which algorithmic management implies new parameters of the human experience of work. As
   Möhlmann et al. (2021) point out, algorithmic systems that can drive improved fairness and
   consistency are ‘black boxes’ that offer little transparency or recourse to the employees they
   govern. Workers may lose agency if the actions on workload, breaks or performance ratings are
   taken without being dictated by human dialogue or contextual knowledge (Ivanova et al., 2018).
   But as output increases or task performance improves measured by this method, the stress can
   increase, job satisfaction and burnout can also increase (Tarafdar et al., 2019; Wood et al., 2019).
   Such a contradictory outcome is the underlying concept of the algorithmic management
   dilemma—a concept suggesting a dual effect: the performance gain is accompanied with
   psychological costs.
   Initial framings of algorithmic tools were within the realm of technological optimism because the
   tools could reduce managerial bias and increase productivity (Binns et al., 2018). In fact,
   algorithms are able to provide data rich feedback, predictive analytics and real time corrections
   that could theoretically optimize completion of the task and minimize inefficiencies (Jarrahi et

al., 2021). For example, in warehouse settings algorithmic scheduling has reduced idle time and
improved throughput (Rosenblat & Stark, 2016). However, while the same systems might
champion metrics and micro surveillance and continuous assessment they may threaten
creativity, adaptability and emotional resilience (Veen et al., 2020). Particularly sharp are gig
worker reports of feeling ‘trapped in metrics’, i.e., in the sense that all their actions are under
surveillance and measured against opaque performance benchmarks (Moore, Upchurch, &
Whittaker, 2018). The psychological consequences of these are consistent with the technostress
framework (Tarafdar et al., 2019) that frames digital technologies as both an enabler as well as a
stressor based on the extent of control exercised by the user, training and support systems
associated with these technologies.
Concerns have also been raised in regards to the dehumanizing components of AM by scholars.
Although traditional supervisors may be empathetic, judicious and make contextual judgments,
algorithmic managers inflexibly and lacking emotional nuance impose rules (ParentRocheleau &
Parker, 2021). Feedback and decision making without interaction with people erodes relational
trust leading to feelings of alienation or depersonalized work (Kellogg et al., 2020). Additionally,
employees are frequently left out of the design and iterative process for these systems and, as a
result, are left powerless to participate in and bargain collectively (Galière, 2022). It adds to the
ethics of digital labor rights, algorithm accountability and workplace fairness particularly in
settings in which the workforce is already precarious.
As such, AM is a dual technical and socio-technical phenomenon and the discussion regarding
its implications for STS theory resonates with broader debates regarding the nature of
technological interventions and their direct instrumental and indirect socio-humanistic impacts
(Trist & Bamforth, 1951). Rather, AM tools are not neutral, but rather form and are formed by
the organizational environment (i.e., culture, labour relations and institutional frameworks) in
which they are situated (Zuboff, 2019). An alternative lens to pursue this inquiry is similarly
provided by the Job Demands-Resources (JD-R) model. According to Bakker and Demerouti
(2007), job demands, e.g., surveillance and inflexible metrics, deplete energy and psychological
resources; conversely job resources, like feedback and autonomy, enhance engagement and well
being. Employee strain ensues and work satisfaction declines as the number of personal

resources and work demands rises, but algorithmic management results in poor resource
allocation.
These tensions need to be addressed and thus, we need to empirically gauge the effects of
algorithmic management on task performance and employee well being. Despite the surge of a
significant body of work that has focused either on performance gains or stress outcomes and
within different employment contexts, a paucity of studies considers both performance gains and
stress outcomes together. In order to address this gap, this study combines quantitative
performance metrics with qualitative insights on worker experiences to analyze the double-edged
sword effect of AM. The aim of doing so is to take part in the development of algorithmic
governance frameworks that are balanced and ethical, while enabling productivity without
compromising human dignity of labor.

2. Literature Review
   Algorithmic management (AM) is changing organizational hierarchies, labor processes and the
   psyche of labor. Management authority and decision making were traditionally resident in direct
   human supervision where judgment was discretionary and negotiated between people. In
   digitized workplaces though, where I will focus my analysis, many of these responsibilities have
   been de-politicized and shifted to algorithmic systems that draw upon big data, predictive
   analytics and real time monitoring; which then allocate tasks, assess performance and enforce
   compliance (Ajunwa, 2020). It is particularly deep in platform labor, including food delivery,
   ridesharing and freelance marketplaces and is now spreading deep into traditional corporate
   activities (Bajwa et al., 2021). Although this ambition towards greater efficiency and impartiality
   stands to bring advantages, the literature is increasing its documentation on the potentially
   harmful results that this evolution can have for the emotional, psychological and social well
   being of workers.
   The opacity and perceived fairness of algorithmic systems are one of the key debates in the
   literature. Algorithms do not usually explain or have flexibility about context when enforcing
   decisions (Pasquale, 2015). Because algorithmic logic is of a “black box” nature, it creates a
   feeling of randomness and powerlessness for the workers, destroying the typical organizational
   trust (Eubanks, 2018). Angrave et al. (2016) have also focused on how opaque performance

analytics make workers more anxious, outraged and less satisfied with their jobs when they are
not allowed to understand or influence their opaque metrics. The result of this produces what
O’Neill (2016) calls ‘weapons of math destruction’, utilizing flawed or biased data to facilitate
unfair outcomes that are at a high level hard to refute.
Furthermore, literature has demonstrated that algorithmic control increases work intensification
and emotional labor to unsustainable degrees. According to Fleming (2017), algorithmic
management replicates a kind of ‘digital Taylorism’: the minutia of an employee’s work
becomes monitored and quantified and overseen in real time. In this condition, workers are
expected (although unwillingly) to carry out the performative labor, described by Diefenbach
and Sillince (2011), as working tirelessly to optimise ‘visibility’ and ‘productivity’, regardless of
the loss of authenticity and health. Leighton (2019) documents, for example, how Amazon
warehouse employees were pushed by algorithmic weights to forgo bathroom breaks, to
downplay injuries and to stay in an escalated state of productivity at the expense of burnout and
attrition.
On the other hand, algorithmic systems also provide opportunities to reduce bias and standardize.
In one vein of research, AM tools can decrease favoritism and discrimination by ensuring
uniformity by utilizing well designed AM tools to enforce uniform rules and performance
standards (Binns, 2019). For example, algorithms used in HR analytics can increase diversity by
reducing unconscious bias during candidate, evaluation and resume screening, according to Kim
(2017). But even these benefits rely on knowing that these systems are trained with data that is
transparent and fair. However, if the training data contain past inequalities or discriminatory
hiring patterns, the algorithm will replicate and perfect these patterns, as reported by Noble
(2018) in his study on algorithms bias in digital platforms.
The literature in the domain of employee autonomy takes a nuanced view. However, other
scholars contend that, far from diminishing agency, AM forces conformity to rigid behavioral
scripts (Scholz, 2017); and yet still other scholars argue that algorithmic tools can enhance
autonomy in specific settings. Boudreau and Lakhani (2013) describe how task automation and
data driven scheduling can let complete mundane decisions, freeing up employees to focus on
their strategic or creative work. But the benefit of AM is often only afforded to higher skilled

professionals and low wage or precarious workers benefit from AM as a mechanism of control.
Turn to Dubai (2020), who notes the contrast between the experience of gig drivers, on the one
hand, with remote software engineers, on the other, who both work under algorithmic systems
but in significantly different ways.
Recent studies concerning algorithmic governance have also investigated the psychological
aspect of worker relations with the employer in terms of psychological contract. Workers’
understandings about the reciprocal obligations that they believe exist between themselves and
their organization are denoted in the psychological contract (Rousseau, 2001). The perceived
relational component of work may erode as algorithmic systems take control of performance
evaluation, reward and feedback; such systems can operate with a transactional model where the
only significant factor is outcome. However, this can result in disengagement if required in roles
which have to be carried out with intrinsic motivation and emotional investment (Van den Broek
et al., 2021). In addition, relegating the feedback and disciplinary functionalities to the machines
may generate reduced accountability since workers do not know to whom or where to attribute
blame or even seek solution to the adverse outcome (Elish, 2019).
There is also literature on organizational justice on the effects of AM. The dimensions of
organizational justice include fairness of outcomes (distributive), fairness of the process
(procedural) and fairness in interpersonal treatment (interactional) (Colquitt, 2001). Algorithms
can arguably improve distributive and procedural fairness by setting standardized rules, but by
doing so may diminish interactional justice by removing what is arguably completely human
about management interaction. Faraj et al. (2018) have shown that workers also value critical
interpersonal feedback as a mechanism for clarification, empathy and mutual adjustment, all of
which are functions that algorithmic tools do not yet replicate.
Finally, scholars have begun to consider how algorithmic management enables new forms of
resistance and labor mobilization. However, workers are increasingly creating
counter–algorithmic forms of resistance (Rosenblat 2018),including gaming the metrics,
organizing digitally and exercising legal rights. For example, there are reports that Uber drivers
cheat smartphone GPS location or hold rides long before users reach destinations just to
artificially raise their ratings in the Uber algorithms (Gray & Suri 2019) and platform based

worker’s unions are fighting against algorithmic opaqueness and for having rights to contest
ratings (Gray & Suri 2019). They signify that AM does more than reshape managerial control: it
reconfigures the geography of worker agency and collective action.
Overall, the literature paints a mixed and often conflicting picture of algorithmic management.
Although it promises potential benefits of efficiency, standardization and dependability, it is
dangerous in its ability to alienate workers, extinguish freedom and intrude on well-being. AM
may be affected by design choices, contextual variables and power dynamics, not solely as a
monolith. To continue to understand these nuanced effects, we need to continue to engage in
empirical inquiry of how algorithmic systems can be designed efficiently and ethically aligned
with human centered work.

3. Methodology
   3.1 Research Design
   The research was based on a quantitative, cross-sectional survey design and examined the double
   edged impact of algorithmic management on employee well being and task performance.
   However, since the survey approach is appropriate for collection of data among a wide
   population, for measurement of subjective constructs such as perceived well-being and perceived
   autonomy and is cost effective for spatial dispersion of subjects, the survey approach was
   selected. The main objectives were to record self reported perceptions and experiences of
   individuals working within algorithm enabled systems of work and to explore the relationship
   between algorithmic management, psychological well being and individuals task performance.
   3.2 Population and Sampling Strategy
   This study targeted employees of sectors that make use of algorithmic management tools:
   logistics, customer support, food delivery, e-commerce and technology enabled services. Both
   full time workers and gig-based or flexible workers ensured quality representation across
   employment categories. Non probability purposive sampling technique was used focussing on
   people with direct exposure of algorithmic systems like automated task scheduling, real time
   performance dashboards and algorithmic feedback or penalty mechanisms. A total of 312

respondents from across Pakistan, India, the UAE and select regions of Europe and North
America make up the final sample. All respondents confirmed their work in algorithmically
mediated work environments prior to participation.
3.3 Instrumentation and Survey Design
Four structured sections made up the survey instrument. The demographic data gathered
included age, gender, education level, job type and industry sector. The second section measured
algorithmic management exposure using a 7 item custom developed scale based upon the
presence of features like automated task assignment, performance monitoring and feedback
through algorithmic dashboards. Some existing works on digital labor platforms were adapted
and validated as items.
The third section measured employees well‐being using items from the General Health
Questionnaire (GHQ‐12) and the Work‐Related Quality of Life Scale (WRQoL). For example,
subjective feelings of autonomy, perceived surveillance, stress, emotional exhaustion and the
ability to disengage from work were all subjective. The fourth and last section reported task
performance, both in terms of perceived self-efficacy and also as comparative performance (e.g.
“I now complete tasks more efficiently than before algorithmic tools were introduced”) using a
validated, adapted scale from Williams and Anderson’s (1991) in-role performance instrument.
The measuring scale for all items was a 5 point Likert scale ranging from “Strongly disagree” (1)
to “Strongly agree” (5).
3.4 Pilot Testing and Validation
The survey was piloted and tested face validity and reliability with 30 respondents before
fielding. Using feedback, ambiguous questions were reworded and redundant items were
dropped. All subscales were found to have high internal consistency: algorithmic management
exposure (α = 0.83); well-being (α = 0.89); and task performance (α = 0.87). Given that
exploratory factor analysis (EFA) confirmed the anticipated factor structure for all three
constructs, construct validity was evaluated.
3.5 Data Collection Procedure

Over five weeks the finalized survey was delivered using online channels, namely Google Forms
and Typeform. The participation was totally voluntary and anonymous. Before the survey, a
consent statement was made indicating the purpose of the study, making participation voluntary,
the confidentiality of data and information regarding ethical clearance. To improve response
rates, reminder emails were sent and social media posts were used and no financial incentives
were offered. Individuals were free to withdraw from the study at any time before final
submission.
3.6 Data Analysis Techniques

Statistical analysis was done on data that were exported to SPSS version 27. Descriptive
statistics for demographic components and the distribution of responses across variables are
presented. The relationships among algorithmic exposure, well being indicators and performance
scores were assessed by using Pearson correlation analysis. Secondly, multiple linear regression
models were developed to assess how much variance in task performance and well being could
be predicted by algorithmic management, accounting for demographic variables. Furthermore,
PROCESS macro (Model 4) was used to do mediation analysis to see whether there exists the
mediating role of (perceived) autonomy between the effect of algorithmic control and
psychological well-being. Significance level was p < 0.05 for all tests.
3.7 Ethical Considerations
The ethical standards of the Declaration of Helsinki and approval from the university’s
Institutional Review Board (IRB) were followed in this study. The study objectives, right to
withdraw at any time and storage and use of responses were explained to participants. The data
was strictly confidential and used only for academic research. All statistical reporting was in
aggregate form and no identifying information was collected.

4. Results

4.1 Demographic Overview
For understanding the profile of respondents, demographic data were analyzed and presented in
Table 1 and Figures 1 . Our sample (N = 312) was almost evenly split across genders: 162 males
(51.9%) and 150 males (48.1%) as shown in Donut Chart (Figure 1). As far as the age was
concerned, the lollipop chart (Figure 2) showed the largest age percentile trending between
26–35 (44.2%), followed by 18–25 (30.4%) and 36–50 (25.3%). Variety in employment types
saw 57.7% of them being employed full time, while the remaining 42.3% were employed on gig
and contract terms. The industry was broadly represented in the participants who were from
logistics, customer service and e-commerce sectors. The key demographic insights provided a
basis for understanding the impact of Algorithmic Management on a wide-ranging set of
worker’s profiles.
Table 1. Demographic Profile of Respondents (N = 312)
Category Value Frequency Percentage (%)
Gender Male 162 51.9
Female 150 48.1
Age Group 18–25 95 30.4
26–35 138 44.2
36–50 79 25.3
Employment Type Full-Time 180 57.7
Gig/Contract 132 42.3
Sector Logistics 102 32.7
Customer Service 108 34.6
E-Commerce 102 32.7

Figure 1 Age Group Distribution

4.2 Descriptive Statistics and Reliability
The descriptive statistics of the three main constructs: Algorithmic Management Exposure
(AME), Employee Well-Being (EWB) and Task Performance (TP), as well as their internal
consistency metrics are presented in Table 2. In Figure 2, the radar chart above shows the means
visually by stating that task performance had the highest mean score (M = 4.15, SD = 0.67),
followed by algorithmic exposure (M = 3.91, SD = 0.72). Relatedly, employee well-being was
significantly lower (M = 2.83, SD = 0.85), indicating a possible inverse relationship between
algorithmic control and psychological outcomes. The Cronbach’s alpha values for all three
constructs were strong and above the threshold value of 0.80.
Table 2. Descriptive Statistics of Main Variables
Variable Mean Standard Deviation Min Max Cronbach Alpha

Algorithmic Management Exposure 3.91 0.72 1.0 5.0 0.83

Employee Well-Being 2.83 0.85 1.0 5.0 0.89
Task Performance 4.15 0.67 2.0 5.0 0.87
Figure 2: Radar Chart of Main Variable Means

4.3 Algorithmic Exposure Trends
Table 3 presents detailed item-wise responses for algorithmic exposure and a stacked bar chart
(Figure 3 ) visualizes them’. A majority of respondents (71%) reported they are subject to
algorithmic systems for task allocation, (67%) to real-time monitoring and (69%) to automated
feedback mechanisms. The majority (nearly 60%) indicated that they had received
algorithmically driven penalties and more than 70% were aware that visibility and ranking were
impacted by dashboards. The exposure confirmed that the AM systems are deeply engaged at the
work routine of the respondents.
Table 3. Frequency Distribution of Algorithmic Exposure Items
Item Strongly Disagree Disagree Neutral Agree Strongly Agree
Automated task allocation 12 28 49 131 92

Real-time monitoring 10 34 57 120 91
Algorithmic feedback 8 26 61 124 93
Automated penalties 25 40 58 102 87
Performance ranking system 18 31 52 115 96
Predictive task scheduling 22 33 50 110 97
Dashboard control visibility 15 29 48 120 100
Figure 3 Distribution of Algorithmic Exposure Items

4.4 Employee Well-Being Patterns
Finally, as illustrated in Figure 4 visualizing Table 4, the psychological impact from algorithmic
management is shown to be very negative. A high proportion of respondents agreed with
statements of emotional exhaustion (63%) and persistent work related stress (62%). They also
had a sense of lack of control over their schedules (56%) and worried about performance
evaluations (63%). However, fascinatingly, in contrast to enjoyment of daily tasks, only 52
percent agreed or strongly agreed with positive sentiments which implies a lack of psychological

buffer against stress. The diverging visualization shows how inflated the perception is in
polarity, while the emotional stress caused by AM systems is shown.
Table 4. Frequency Distribution of Employee Well-Being Items
Item Strongly
Disagree

Disagree Neutral Agree Strongly
Agree

Feel mentally exhausted after
work

15 38 62 120 77

Lack of control over daily
schedule

22 43 71 105 71

Work-related stress persists after
hours

18 36 65 118 75

Feel anxious about performance
ratings

16 30 70 128 68

Enjoyment in daily tasks 28 48 74 100 62
Figure 4 Distribution of Well-Being Items

4.5 Task Performance Perceptions
Participant responses to performance related items are outlined in Table 5 with a slope chart
clarifying responses shown in Figure 5. More than 75 percent of respondents agreed or strongly
agreed that they get work done efficiently and meet performance goals. Timely feedback; task
confidence; and self reported improvements in productivity, all revealed high Scores. This work
is consistent with literature that shows that algorithmic feedback loops can improve observable
task performance. When put together with well-being scores, however, a paradox is suggested:
people perform better, even as they subjectively deteriorate, supporting the much talked about
‘double edged sword’ hypothesis of algorithmic management.
Table 5. Frequency Distribution of Task Performance Items
Item Strongly
Disagree

Disagree Neutral Agree Strongly
Agree
Complete tasks more efficiently 5 14 50 140 103
Meet performance targets
consistently

4 12 47 145 104

Feel confident in task execution 6 18 42 138 108
Receive timely performance
feedback

3 13 41 139 116

Productivity has improved with
AM tools

2 11 45 142 112

Figure 5 Distribution of Task Performance Items

4.6 Correlation Analysis.
The correlation matrix is summarized in Table 6 and these relationships are visualized using a
cleanly annotated heatmap in Figure 6. The relationship between algorithmic exposure and task
performance was moderately positively correlated (r = 0.61) which supports the claim that AM
has a positive effect on output quality. The negative AM and well being correlation (r = -0.52)
actually means that exposure to automated control mechanisms can severely deteriorate mental
and emotional health. Furthermore, an inverse relationship was found between well being and
task performance (r = -0.38) such that increased performance was associated with decreased
psychological sustainability.
Table 6. Correlation Matrix of Key Variables
Variables Algorithmic Exposure Employee Well-Being Task Performance
Algorithmic Exposure 1.00 -0.52 0.61
Employee Well-Being -0.52 1.00 -0.38

Task Performance 0.61 -0.38 1.00
Figure 6 Correlation Matrix of Key Variables

4.7 Regression Model Insights
A multiple regression model was constructed to investigate predictors of task performance and
the results are presented in Table 7 and depicted in Figure 6 in a horizontal bar chart of
standardized beta weights. Perceived autonomy (β = 0.31, p < 0.001), employee well-being (β =
0.21, p < 0.001) and algorithmic exposure (β = 0.45, p < 0.001) were the strongest predictors.
They confirm that algorithmic systems affect significantly performance outcomes and highlight
the mediating effect of perceived control. The standardized beta chart affords a clear visual
comparative for how algorithmic exposure predominates performance prediction.

Table 7. Regression Model Summary (DV: Task Performance)
Predictor Unstandardized B Standard Error Standardized Beta t-value p-value

Algorithmic
Exposure

0.58 0.08 0.45 7.25 0.000

Employee Well-
Being

0.25 0.06 0.21 4.17 0.000

Perceived Autonomy 0.37 0.07 0.31 5.29 0.000
Figure 7 Standardized Beta Weights (Task Performance Model)

4.8 Mediation Effect of Autonomy
A mediation analysis was attempted to test if perceived autonomy mediates the relationship
between algorithmic exposure and well being. A circular mediation path diagram of the results is
shown in Figure 8 and in Table 8. Both the path from algorithmic exposure to autonomy (β =
0.42, p < 0.001) and the path from autonomy to well-being (β = 0.39, p < 0.001), were
significant. Statistically significant indirect effects were observed (β = -0.15, CI [-0.24; -0.08]),

so partial mediation was confirmed. Interestingly, we find that the direct effect of algorithmic
exposure to well-being remained significant (β = -0.37) indicating that autonomy only partially
explains the effect. The mediation diagram visualizes these interactions, showing the extreme
complexity of employee experience in algorithmically managed environments.
Table 8. Mediation Analysis Summary (Mediator: Autonomy)
Path Effect Size CI Lower CI Upper p-value
A → M (AE → Autonomy) 0.42 0.31 0.55 0.000
M → B (Autonomy → Well-Being) 0.39 0.29 0.48 0.000
Total Effect (AE → WB) -0.52 -0.63 -0.40 0.000
Direct Effect -0.37 -0.47 -0.26 0.000
Indirect Effect -0.15 -0.24 -0.08 0.002
Figure 8 Mediation Model: Autonomy as Mediator

5. Discussion
   This research offers compelling evidence that AM has a double edged impact on task
   performance and employee well being. This paradox fits with theoretical frameworks such as the
   High Performance Work System (HPWS) which theories that technologies created for efficiency
   may lead to increased work intensification and stress (Boxall & Macky, 2009). Quantitative
   results in this study showed positive AM exposure–self reported task performance correlations,
   but qualitative implications in terms of the examined well-being metrics were increased
   psychological strain which aligns with concern in the scholarly literature (Brougham & Haar,
   2018) arguing that unchecked algorithmic logic dehumanizes the worker.
   Control displacement seems to be one of the core mechanisms behind this paradox. As
   algorithms not only begin to influence decision-making processes, like task allocation,
   performance feedback or even disciplinary action, workers may even feel a sense of loss of
   agency and relational connection to their employer. According to Rahman (2021) this is a
   phenomenon called ‘algorithmic despotism,’ which deprivates employees of the ability to
   negotiate working terms or a sign of empathetic understanding during their performance
   disputes. Similarly, Foucauldian interpretations of surveillance are in tune with workers
   internalizing algorithmic expectation as well as self – monitoring on a continual basis which can
   create burnout and erode morale (Zuboff, 2015).
   In addition, the findings demonstrate that perceived autonomy mediates the relation between
   algorithmic exposure and well being that is consistent with Self Determination Theory (SDT)
   that states the necessity of autonomy, competence and relatedness (Deci and Ryan, 2000). When
   such algorithmic systems determine break times, pace and performance metrics, these
   psychological needs are not being satisfied which tends to decrease motivation and satisfaction.
   Previous research suggests that lack of autonomy under AM results in low levels of job
   embeddedness and high levels of turnover intentions (Cheng & Foley, 2019). Mediation analysis
   conducted in this study empirically verifies the partial buffer effect of autonomy which provides
   empirical evidence that AM systems permitting some level of user input or override can offset
   harm to employee wellbeing.

This is consistent with other AM and digital operations and logistics research which shows the
use of algorithm tools can improve output consistency and decrease idle times (Ghosh 2022).
Yet, many of these productivity gains involve cognitive overload and role conflict where
algorithms signal contradictions or are opaque in how they decide (Schildt, 2017). In particular,
algorithmic ratings and penalties can contribute to chronic anxiety among workers, as is notable
on platform based gig work (Anwar and Graham, 2020). For example, overly rigid performance
dashboards promote short term performance spikes that trade long term engagement and
innovation even within formal employment contexts (Faraj, Pachadi, & Sayegh, 2018).
This study finds a strong negative correlation between algorithmic exposure and well‐being, a
result consistent with recent research in AI‐mediated workplaces: widespread reporting on the
emotional cost of digital micromanagement. For example, for example, ‘technostress creators’
like technostress invasion and technostress overload have burst forth within (and especially jobs
that experience brutal, relentless and unforgiving algorithmic feedback. As a result, there are
calls for ‘algorithmic resilience’ (Tarafdar et al., 2020) which advocates for protective
frameworks for workers and organisations to be able to cope with continuous digital governance.
A second notable feature of the study is a strong but moderate inverse relationship between well
being and task performance which implies that workers can remain highly productive with
algorithmic control, even as their emotional strain worsens. The idea of surface acting from the
emotional labor literature—that is, employees should suppress emotions to match the
expectations of the organization (Grandey, 2000)—is supported. Though this approach is likely
to generate short term results, in the long run it is unsound, feels emotionally exhausting and
reduction in organizational citizenship behaviors tend to take place (Cropanzano, Rupp, &
Byrne, 2003).
Additionally, the effects of AM in context deserve attention. Prior research has well documented
the impact of AM in the gig economy, but this research extends it to full time employees in
logistics, customer service, as well as from e-commerce. Furthermore, by expanding the scope,
we understand that AM does not only apply to platform labor (Mateescu, 2021), but it is quickly
penetrating traditional organizational structures. This expansion requires human centred design

principles and technological innovation to be developed within multi level governance
frameworks (Wajcman, 2020).
They also suggest policy and managerial implications. Efficiency driven automation is not
autonomous operation in the vacuum, rather efficiency driven automation directly affects worker
psychology and behavior. Based on the harms identified in this study, companies should design
AM systems that integrate ethical design principles such as transparency, contestability and
hybrid feedback models that have both algorithmic and human inputs to offset these harms. Says
Mittelstadt et al. (2016): Algorithmic accountability is not just a technological requirement but
an ethical imperative. The European Commission (2021) even now has started drafting the EU
AI regulations that focus on human oversight of algorithmic decision making among the
institutions. These regulatory movements may represent a model for what organizations need to
do in order to implement AM without sacrificing worker dignity.
Overall, this paper joins the growing literature considering algorithmic management as much a
social actor constructing power, performance and wellbeing in contemporary workplaces as it is
a technological tool. The research underscores the importance of striking a balance between
efficiency and empathy, as well as automation and autonomy when designing the future of work.

References
● Bakker, A. B., & Demerouti, E. (2007). The Job Demands‐Resources model: State of
the art. Journal of Managerial Psychology, 22(3), 309–328.
https://doi.org/10.1108/02683940710733115

● Binns, R., Veale, M., Van Kleek, M., & Shadbolt, N. (2018). ‘It’s reducing a human being
to a percentage’: Perceptions of justice in algorithmic decisions. In Proceedings of the
2018 CHI Conference on Human Factors in Computing Systems (pp. 1–14). ACM.
https://doi.org/10.1145/3173574.3173951

● Galière, S. (2022). Governing workers through algorithmic management: The case of
Deliveroo in France. Work, Employment and Society, 36(5), 890–906.
https://doi.org/10.1177/09500170211038668

● Ivanova, M., Bronowicka, J., Kocher, E., & Staab, P. (2018). The App as a Boss? Control
and Autonomy in Application-Based Management. Hans-Böckler-Stiftung.

● Jarrahi, M. H., Sutherland, W., Nelson, S. B., & Sawyer, S. (2021). Algorithmic
management in a work context. Business Horizons, 64(4), 477–486.
https://doi.org/10.1016/j.bushor.2021.02.019

● Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new
contested terrain of control. Academy of Management Annals, 14(1), 366–410.
https://doi.org/10.5465/annals.2018.0174

● Lee, M. K., Kusbit, D., Metsky, E., & Dabbish, L. (2015). Working with machines: The
impact of algorithmic and data-driven management on human workers. In Proceedings of
the 33rd Annual ACM Conference on Human Factors in Computing Systems (pp.
1603–1612). https://doi.org/10.1145/2702123.2702548

● Mateescu, A., & Nguyen, A. (2019). Algorithmic Management in the Workplace. Data &
Society Research Institute.

● Möhlmann, M., Zalmanson, L., Henfridsson, O., & Gregory, R. W. (2021). Algorithmic
management of work on online labor platforms: When matching meets control. Academy
of Management Discoveries, 7(4), 560–583. https://doi.org/10.5465/amd.2019.0284

● Moore, P. V., Upchurch, M., & Whittaker, X. (2018). Humans and Machines at Work:
Monitoring, Surveillance and Automation in Contemporary Capitalism. Springer.
https://doi.org/10.1007/978-3-319-63257-5

● Newlands, G. (2021). Algorithmic surveillance in the gig economy: The organization of
work through Lefebvrian conceived space. Organization Studies, 42(5), 759–777.
https://doi.org/10.1177/0170840620946829

● Parent-Rocheleau, X., & Parker, S. K. (2021). Algorithms as work designers: How
algorithmic management influences the design of jobs. Human Resource Management
Review, 31(4), 100696. https://doi.org/10.1016/j.hrmr.2021.100696

● Rosenblat, A., & Stark, L. (2016). Algorithmic labor and information asymmetries: A
case study of Uber’s drivers. International Journal of Communication, 10, 3758–3784.

● Tarafdar, M., Cooper, C. L., & Stitch, J. F. (2019). The technostress trifecta—Techno
eustress, techno distress and design: Theoretical directions and an agenda for research.
Information Systems Journal, 29(1), 6–42. https://doi.org/10.1111/isj.12169

● Trist, E. L., & Bamforth, K. W. (1951). Some social and psychological consequences of
the longwall method of coal-getting. Human Relations, 4(1), 3–38.
https://doi.org/10.1177/001872675100400101

● Veen, A., Barratt, T., & Goods, C. (2020). Platform-capital’s ‘app-etite’ for control: A
labour process analysis of food-delivery work in Australia. Work, Employment and
Society, 34(3), 388–406. https://doi.org/10.1177/0950017019836911

● Wood, A. J., Graham, M., Lehdonvirta, V., & Hjorth, I. (2019). Good gig, bad gig:
Autonomy and algorithmic control in the global gig economy. Work, Employment and
Society, 33(1), 56–75. https://doi.org/10.1177/0950017018785616

● Zuboff, S. (2019). The Age of Surveillance Capitalism: The Fight for a Human Future at
the New Frontier of Power. PublicAffairs.
● Ajunwa, I. (2020). The quantified worker: Law and technology in the modern workplace.
Cambridge University Press. https://doi.org/10.1017/9781108692889

● Angrave, D., Charlwood, A., Kirkpatrick, I., Lawrence, M., & Stuart, M. (2016). HR and
analytics: Why HR is set to fail the big data challenge. Human Resource Management
Journal, 26(1), 1–11. https://doi.org/10.1111/1748-8583.12090

● Bajwa, U., Knorr, L., Ruggiero, E. D., Gastaldo, D., & Zendel, A. (2021). Towards an
understanding of workers’ experiences in the platform economy: The case of Uber drivers
in Canada. Canadian Journal of Work and Society, 1(2), 55–75.

● Binns, R. (2019). On the apparent conflict between individual and group fairness. In
Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency (pp.
514–524). ACM. https://doi.org/10.1145/3287560.3287592

● Boudreau, K. J., & Lakhani, K. R. (2013). Using the crowd as an innovation partner.
Harvard Business Review, 91(4), 60–69.

● Colquitt, J. A. (2001). On the dimensionality of organizational justice: A construct
validation of a measure. Journal of Applied Psychology, 86(3), 386–400.
https://doi.org/10.1037/0021-9010.86.3.386

● Diefenbach, T., & Sillince, J. A. A. (2011). Formal and informal hierarchy in different
types of organization. Organization Studies, 32(11), 1515–1537.
https://doi.org/10.1177/0170840611421254

● Dubal, V. B. (2020). The time politics of gig work. California Law Review, 105(1),
123–165. https://doi.org/10.15779/Z38ZG6G83S

● Elish, M. C. (2019). Moral crumple zones: Cautionary tales in human-robot interaction.
Engaging Science, Technology, and Society, 5, 40–60.
https://doi.org/10.17351/ests2019.261

● Eubanks, V. (2018). Automating inequality: How high-tech tools profile, police, and
punish the poor. St. Martin’s Press.

● Faraj, S., Pachadi, S., & Sayegh, K. (2018). Working and organizing in the age of the
learning algorithm. Information and Organization, 28(1), 62–70.
https://doi.org/10.1016/j.infoandorg.2018.02.005

● Fleming, P. (2017). The death of Homo economicus: Work, debt and the myth of endless
accumulation. Pluto Press.

● Gray, M. L., & Suri, S. (2019). Ghost work: How to stop Silicon Valley from building a
new global underclass. Houghton Mifflin Harcourt.

● Kim, P. T. (2017). Data-driven discrimination at work. William & Mary Law Review,
58(3), 857–936.

● Leighton, P. (2019). Amazon and the algorithmic management of labour: Management by
metrics and the regulation of work. Journal of Australian Political Economy, 84, 25–45.

● Noble, S. U. (2018). Algorithms of oppression: How search engines reinforce racism.
NYU Press.

● O’Neil, C. (2016). Weapons of math destruction: How big data increases inequality and
threatens democracy. Crown Publishing Group.

● Pasquale, F. (2015). The black box society: The secret algorithms that control money and
information. Harvard University Press.

● Rosenblat, A. (2018). Uberland: How algorithms are rewriting the rules of work.
University of California Press.

● Rousseau, D. M. (2001). Schema, promise and mutuality: The building blocks of the
psychological contract. Journal of Occupational and Organizational Psychology, 74(4),
511–541. https://doi.org/10.1348/096317901167505

● Scholz, T. (2017). Uberworked and underpaid: How workers are disrupting the digital
economy. Polity Press.

● Van den Broek, D., Keating, E., & Reilly, A. (2021). Algorithmic management, work and
employment: The frontier of the new digital workplace. Labour & Industry: A Journal of
the Social and Economic Relations of Work, 31(3), 219–226.
https://doi.org/10.1080/10301763.2021.1939433

● Anwar, M. A., & Graham, M. (2020). Between a rock and a hard place: Freedom,
flexibility, precarity and vulnerability in the gig economy in Africa. Competition &
Change, 24(3-4), 379–403.

● Boxall, P., & Macky, K. (2009). Research and theory on high-performance work systems:
Progressing the high-involvement stream. Human Resource Management Journal, 19(1),
3–23.

● Brougham, D., & Haar, J. (2018). Smart technology, artificial intelligence, robotics, and
algorithms (STARA): Employees’ perceptions of our future workplace. Journal of
Management & Organization, 24(2), 239–257.

● Cheng, C. F., & Foley, S. (2019). Algorithmic management and employee voice: The role
of control and perceived fairness. Journal of Business Ethics, 154(1), 123–135.

● Cropanzano, R., Rupp, D. E., & Byrne, Z. S. (2003). The relationship of emotional
exhaustion to work attitudes, job performance, and organizational citizenship behaviors.
Journal of Applied Psychology, 88(1), 160–169.

● Deci, E. L., & Ryan, R. M. (2000). The “what” and “why” of goal pursuits: Human needs
and the self-determination of behavior. Psychological Inquiry, 11(4), 227–268.

● European Commission. (2021). Proposal for a regulation laying down harmonised rules
on artificial intelligence (Artificial Intelligence Act). Brussels.

● Faraj, S., Pachadi, S., & Sayegh, K. (2018). Working and organizing in the age of the
learning algorithm. Information and Organization, 28(1), 62–70.

● Ghosh, A. (2022). Performance optimization in algorithmic retailing: A data analytics
perspective. International Journal of Retail & Distribution Management, 50(2), 115–130.

● Grandey, A. A. (2000). Emotion regulation in the workplace: A new way to
conceptualize emotional labor. Journal of Occupational Health Psychology, 5(1),
95–110.

● Mateescu, A. (2021). Algorithmic management in traditional workplaces: Emerging
research and questions. Data & Society Research Institute.

● Mittelstadt, B. D., Allo, P., Taddeo, M., Wachter, S., & Floridi, L. (2016). The ethics of
algorithms: Mapping the debate. Big Data & Society, 3(2), 1–21.

● Moore, P. V. (2021). Surveillance, stress and technostress: An occupational health
perspective. Surveillance & Society, 19(1), 71–75.

● Rahman, H. Z. (2021). Algorithmic despotism and digital Taylorism in the platform
economy. Work, Employment and Society, 35(6), 1091–1108.

● Schildt, H. (2017). Big data and organizational design–the brave new world of
algorithmic management and computer augmented transparency. Innovation:
Organization & Management, 19(1), 23–30.

● Tarafdar, M., Pullins, E. B., & Ragu-Nathan, T. S. (2020). Technostress: Negative effect
on performance and possible mitigations. Information Systems Journal, 30(2), 303–332.

● Wajcman, J. (2020). The digital architecture of time management. Science, Technology,
& Human Values, 45(5), 1007–1028.

● Zuboff, S. (2015). Big other: Surveillance capitalism and the prospects of an information
civilization. Journal of Information Technology, 30(1), 75–89.