Algorithmic precarity and metric power: Managing the affective measures and customers in the gig economy

Quantification of work through performance metrics has long been a critical feature of labor control in manufacturing and service workplaces (Braverman, 1974; Moore, 2018a; Ranganathan and Benson, 2020). Scholars have shown that corporations rely upon algorithms and data to quantify and evaluate workers’ performance in real-time (Kellogg et al., 2020; Moore, 2018a). The platform-mediated gig economy—which is “characterized by independent contracting that happens through, via, and on digital platforms” (Woodcock and Graham, 2020: 11)—amplifies the trend of quantification through the reliance on algorithmic technologies for managing workers at a distance (Jarrahi et al., 2021; Kellogg et al., 2020). One notable feature of algorithmic management is the integration of metrics, such as customer-sourced ratings and system-generated behavioral measures, into algorithmic evaluation and discipline (Gandini, 2019; Kellogg et al., 2020; Stark and Pais, 2021). An Uber driver quoted in a technology blog article, for example, shared, “We’re not just working for money … We’re working for ratings, but ratings have no value. Ratings serve only to prevent you from getting fired” (Dzieza, 2015: para 27). A minor change in customer-sourced ratings can hinder workers’ job prospects, though workers have limited ability to dispute the influences of ratings (e.g. Chan, 2019b; Rahman, 2021; Rosenblat, 2018). The fragility and opacity of ratings contribute to workers’ feelings of anxiety and insecurity (Kellogg et al., 2020; Ravenelle, 2019; Wood and Lehdonvirta, forthcoming), or what Duffy (2020) terms “algorithmic precarity.”

Much scholarship has examined the disciplinary outcomes of metrics on various geographically tethered (e.g. ride-hailing and domestic services) and cloudwork (e.g. online freelancing) platforms (Griesbach et al., 2019; Sutherland et al., 2020; Ticona and Mateescu, 2018; Veen et al., 2020). This comparative research examines how the affective power of metrics and workers’ precarious experiences vary across geographically tethered platforms—including ride-hailing, delivery, and domestic services platforms—in the United States. Geographically tethered platforms require workers to perform an on-demand task in a particular location (Woodcock and Graham, 2020). The three cases of gig work involve some forms of service interactions with customers, which situate metrics into the service triangle—a “three-way contest for control and satisfaction” (Leidner, 1993: 22). There are two main types of metrics, namely customer-sourced ratings and system-generated behavioral measures. Customer-sourced ratings entail customers’ assessment of service quality and can impact one's employment opportunities on the platform. On ride-hailing and delivery platforms, customer-sourced ratings can determine the continuity of platform-based employment. On TaskRabbit, ratings affect the visibility of workers’ online profiles. Additionally, system-generated behavioral measures are often designed to facilitate efficient service transactions. Examples include workers’ job acceptance rates, cancelation rates, and speed metrics (e.g. how long a courier takes to deliver, how quickly a TaskRabbit worker responds to a service request, and how fast a ride-hailing driver drives).

I argue that metrics—and more specifically, what metrics make visible, to whom, and for what purposes—serve as a valuable avenue for examining the algorithmic discipline of workers (Kellogg et al., 2020) as they reveal how the measured ought to be represented. Building upon Beer's (2016) work on metric power, I conceptualize metrics as “affective measures” that “are aimed at stimulating anticipation and uncertainty—often coupling these with senses of insecurity and precarity” (p. 210). Metrics provoke preemptive and reactive affective responses (Beer, 2016; Espeland and Sauder, 2016). As we consider the (in)visible metrics as affective objects that have the potential to make workers feel anxious and precarious (Beer, 2016), how do gig workers encounter, feel, and manage metrics across labor platforms? In what ways do metrics shape and intensify gig workers’ precarious experiences? How do platform-based metrics shape the power dynamics between platforms, workers, and customers in the gig economy?

Drawing on a qualitative, multi-case study of ride-hailing, delivery, and domestic services platforms in the United States, I examine how workers anticipate the influences of metrics, live with metrics, and cope with algorithmic precarity. Data for this project include in-depth interviews with 50 gig workers about their efforts to interpret and manage metrics as part of their everyday work practices. The analysis reveals that these workers were anxious about metrics because of threats of job loss and valued job features (e.g. job flexibility). Information asymmetries and the perceived lack of control over customer-sourced ratings further contributed to workers’ precarious experiences. Therefore, interviewees articulated strategies to manage the uncertainty of customer-sourced ratings—and more precisely, the work-related uncertainty created by “difficult customers”—before, during, and after service interactions. Such strategies hoped to safeguard workers’ ratings and, therefore, their employability; however, they could not resolve workers’ concerns and feelings of anxiety. While anxiety is often considered a psychological state of anticipating an uncertain yet threatening future (Ahmed, 2004), it operates as a technique of governing the self across labor platforms (Beer, 2016; Moore, 2018a). Despite the convergent themes about the affective power of metrics and worker's strategies across the three cases, the (in)visibility of metrics, the settings of platform-mediated worker–customer interactions, and workers’ platform dependence contributed to the varying disciplinary power of metrics.

This study makes three contributions. First, following Beer (2016), it contributes to understanding metrics as affective measures by attending to how metrics shape and are shaped by the trilateral relationship between platforms, workers, and customers. It is through the perceived disciplinary outcomes of metrics and unpredictability of customer that metrics can make workers feel and govern themselves. Second, following previous scholarship (Cameron and Rahman, 2022; Pink et al., 2018), the study offers a processual and relational view of how workers strategically manage the power of metrics throughout platform-mediated labor processes. Third, by examining gig workers’ practices around metrics across labor platforms, the study contributes to understanding the convergence and divergence of metric power in the gig economy.

Metrics are connected with the process of quantification, which is “the production and communication of numbers” (Espeland and Stevens, 2008: 8). Sociologists of quantification have demonstrated the profound impacts of metrics on the subjects of evaluation—which can be individuals, organizations, or social fields—and their decision-making processes (Christin, 2020; Espeland and Stevens, 2008; Mennicken and Espeland, 2019). Three interrelated points explicate the social power of metrics in the literature. First, quantification defines what counts by “turning qualities into quantities on a shared metric” (Espeland and Sauder, 2016: 41). Second, metrics establish social hierarchies of worth in terms of what count as socially desirable (Mau, 2019). Uber's 5-star ratings, for example, do not present drivers’ scores on an ordinal scale, but they can prescribe what counts as “high-quality” service and orient drivers’ attention toward such normative standards (Chan, 2019b; Mason, 2019). Third, metrics are reactive, meaning that “people change their behavior in reaction to being evaluated, observed, or measured” (Espeland and Sauder, 2007: 1).

Particularly relevant to this study is the concept of affective measures, which can be understood as a key mechanism of reactivity (Beer, 2016). Metrics, explain Beer, “as deeds, are a part and product of a series of affective practices that have the power to make people feel uncertain, precarious, and anxious” (p. 203). Metrics are affective objects because metrics are used to provoke particular forms of affective responses that favor productivity and market competition through the production of uncertainty. Workers worry about a low customer-source rating primarily because they may fail to realize “the cluster of things that the object promises” (Berlant, 2011: 24), which could be related to competitive advantage and the evasion of deactivated from labor platforms (see Beer, 2016). Metrics generate perpetual pressure (Mau, 2019) for workers always strive to perform better to avoid becoming targets of punishment. This draws attention to the practices that make metrics affective—how anxiety becomes attached to metrics and circulates across the social world (Ahmed, 2004; Beer, 2016).

The power dynamics of metrics in the gig economy are distinct from those in other contexts such as public measures (e.g. Espeland and Sauder, 2007, 2016) and social media metrics (e.g. Christin and Lewis, 2021; Duffy et al., 2021). Compared to public measures like university rankings, algorithmic metrics remain largely opaque in the gig economy (Rahman, 2021; Rosenblat, 2018). This provides opportunities for exploring how workers navigate metrics in the absence of clear evaluative criteria. Following Leidner's (1993) work on the service triangle, the presence of customers adds another layer of uncertainty to workers’ reactivity to metrics because platforms and workers cannot control customers’ service interactions and evaluative practices. Without occupational training, metrics help to script service interactions through evaluation and discipline. An examination of metrics allows us to consider how they mediate conflict and coordination among platforms, workers, and customers.

Different types of metrics come with various disciplinary outcomes in the gig economy. Most of the platforms (i.e. Uber, Lyft, DoorDash, Instacart, and TaskRabbit) I studied deploy a five-star customer rating system. Like social media metrics (Christin and Lewis, 2021; Cotter, 2019), TaskRabbit workers’ ratings are markers of visibility. From participants’ perspectives, the higher their ratings are, the more likely their online profiles will be seen by prospective clients on TaskRabbit. When ride-hailing drivers and couriers’ average customer-sourced ratings fall under a certain location-based threshold, they may receive warnings or even become “deactivated” by the platforms. Meanwhile, ride-hailing, delivery, and domestic services platforms measure workers’ job acceptance rates, cancelation rates, and other behavioral metrics. Following Beer (2016: 77), “not all measures are equal” since “some become more visible, more telling, or more consequential than others” (italics original). By exploring gig workers’ affective encounters and practices with algorithmic metrics at different stages of the labor process (see Cameron and Rahman, 2022), this study helps understand the heterogeneity of metric power and algorithmic management in the gig economy.

The gig economy encompasses geographically tethered work and cloudwork that are matched and mediated via digital platforms (Woodcock and Graham, 2020). Despite the promises of flexibility and autonomy in the gig economy, workers are subject to precarious work conditions and algorithmic management (Kellogg et al., 2020; Ravenelle, 2019; Wood et al., 2019). Precarious work, explained by Kalleberg and Vallas (2018: 1), refers to “work that is uncertain, unstable, and insecure and in which employees bear the risks of work … and receive limited social benefits and statutory protections” (italics original). Precarity entails not only workers’ anxiety about job loss but also the loss of valued job features, such as task discretion in uncertain futures (Gallie et al., 2017). Much research has highlighted the precarious labor conditions in the gig economy (e.g. Sun et al., 2021; Van Doorn, 2017). By classifying gig workers as independent contractors, labor platforms may disown employment obligations and shift work-related risks to workers (Gregory, 2021; Van Doorn, 2017). According to Jarrett (2022), the informal employment status of platform workers indicates “an almost total absence of formal mechanisms for providing work and skill reproduction security” (p. 51) as well as occupational safety and health (OSH) (see Gregory, 2021).

Algorithmic management, together with the uncertain working conditions, contributes to gig workers’ precarious experiences (Gregory, 2021; Ravenelle, 2019; Wood and Lehdonvirta, forthcoming; Wood et al., 2019). Metrics represents a key technique of algorithmic management (Gandini, 2019; Kellogg et al., 2020). Workers’ metrics may either influence the continuity of platform-mediated employment (Griesbach et al., 2019; Rosenblat, 2018) or affect the accumulation of reputational capital (Fourcade and Healy, 2017; Sutherland et al., 2020; Ticona and Mateescu, 2018). Customer-sourced ratings can be mobilized as the revival of customer control (Cameron and Rahman, 2022; Maffie, 2022; Rosenblat, 2018) and techniques of gamification that attempt to win workers’ consent (Mason, 2019; Vasudevan and Chan, 2022). While reputation systems rooted in interpersonal relations have traditionally been utilized to gain trust in freelance markets, Wood and Lehonvirta (forthcoming) contend that labor platforms contribute to reputational insecurities by disrupting reputation, amplifying the power of customers over workers, and facilitating algorithmic opacity.

What is at stake here is how workers feel algorithms in their everyday encounters with metrics (Beer, 2016; Bucher, 2017; Pink et al., 2018). As Bucher (2017) contends, “what people experience is not the mathematical recipe as such but, rather the moods, affects, and sensations that the algorithm helps to generate” (p. 32). While precarity is often cast as a convergent outcome of algorithmic management, Schor et al. (2020) have demonstrated that economic dependence and barriers for platform participation (e.g. asset requirements) result in variations in precarity (see also Ravenelle, 2019). As precarity can be unequally distributed (Jokinen, 2016), so can feelings of anxiety.

Gig workers can reclaim their labor agency in their everyday lives by learning to navigate metrics—and more broadly, algorithmic management—on which they are being tracked, evaluated, and disciplined. Such everyday practices include unpaid labor, emotional labor, gaming algorithmic rules, and collective sensemaking practices (Cameron and Rahman, 2022; Chan, 2019b; Rosenblat, 2018; Sutherland et al., 2020; Vasudevan and Chan, 2022; Wood and Lehdonvirta, forthcoming). Cameron and Rahman (2022) argue that workers engage in preemptive, interactive, and reactive resistance tactics before, during, and after work. They found that workers’ agency is not evenly distributed throughout the labor process. For example, freelancers can safeguard their ratings by vetting customers before accepting a task, mainly because customers do not know their actions and cannot rate them. There is a diminishing of workers’ latitude to resist the influences of labor control and customers after accepting and performing tasks. Although freelancers might try to gain the favor of customers through interactive tactics during work and submit a complaint about an unfair rating to the platform after work, they have limited control over how customers evaluate their performance and whether the platform would remove the rating. Their processual approach (see also Pink et al., 2018) helps to understand how the affective power of metrics may be circulated at different stages of work.

Accordingly, I examine how gig workers are governed by metrics before, during, and after platform-mediated service interactions. Workers’ anticipatory practices, on the one hand, can be characterized as attempts to maintain their routines and cope with feelings of anxiety. On the other hand, data are continuously produced, so are feelings about data (Pink et al., 2018). Pink et al. (2018) draw attention to how individuals continually cope with data anxieties in their mundane routines of everyday life, whereas Jokinen (2016: 85) argues that precarious everyday agency is “a habit of habit-breaking.” In league with Berlant's (2011) work, Jokinen considers precarization as a social process through which a society becomes more precarious, disorganized, and discontinuous. While habits help individuals to organize their daily life, the repetition of habit-breaking becomes a norm that disorganizes daily routines. To put this simply, Jokinen states, “The sense of the present is intensified in precarious agency. What happened yesterday, or last year, or during half of your life, might not be relevant today” (p. 96). While workers articulate coping strategies to routinize their work practices momentarily, such strategies can be endlessly disrupted by algorithmic management in gig workers’ everyday lives.

This study was undertaken as part of a larger project that investigated how workers encountered, interpreted, and managed algorithmic metrics in the gig economy in the United States. The project began by studying ride-hailing drivers’ (i.e. Uber and Lyft drivers) work practices around ratings,¹ and then included the cases of delivery (i.e. DoorDash, Uber Eats, and Instacart) and domestic services (i.e. TaskRabbit) for comparing the affective power and reactivity of metrics across a range of platform-mediated gig work. Central to a multi-case study approach is “the idea that the objects of investigation are similar enough and separate enough to permit treating them as comparable instances of the same general phenomenon” (Ragin, 1992: 1). The three cases are considered geographically tethered gig work (Woodcock and Graham, 2020) and service work. Workers must complete on-demand tasks and interact with customers in a particular location. The emphasis on service workers helps to situate metrics into specific settings of platform-mediated worker–customer interactions. Meanwhile, the cases vary in terms of the visibility of metrics and metrics-related rewards and sanctions.

Ride-hailing platforms enable drivers and customers to rate one another's performance on a 5-star scale. Drivers’ ratings remain visible to themselves and customers during service interactions, but customers cannot view the former's ratings since then. Drivers’ ratings are not only used for determining the continuity of working on Uber and Lyft but are also one determinant of participating in platform-based rewards programs. Uber Pro rewards program, for instance, requires drivers to maintain at least a 4.85 rating to earn their status in a given period. Additionally, system-generated behavioral measures include acceptance rates, cancelation rates, and speed metrics.

The second case includes workers on Uber Eats, DoorDash, and Instacart. DoorDash and Instacart both use a 5-star rating system, whereas Uber Eats asks customers to give a thumbs up or thumbs down to couriers. Like ride-hailing platforms, couriers’ ratings directly affect the continuity of employment. While DoorDash and Uber Eats match a customer to a driver algorithmically, Instacart uses ratings to determine which delivery orders workers can view and select on the platform. Delivery platforms’ system-generated behavioral measures are similar to that in the case of ride-hailing platforms, but the former (e.g. DoorDash) also measures couriers’ completion rates and on-time/early rates.

In the third case, I examine TaskRabbit, wherein service providers (or “Taskers”) can be hired for various domestic tasks such as moving, cleaning, and furniture assembly. Taskers can communicate with customers before accepting and completing a task. After completing a task, Taskers will be rated on a 5-star scale. TaskRabbit measures workers’ response, acceptance, and reliability rates. A Tasker's profile indicates whether s/he is an “Elite Tasker” and displays individual reviews and a percentage of “positive” ratings (3 to 5 stars). As such, ratings have a high level of visibility for Taskers and their prospective clients, but they are only used for determining the visibility of Taskers’ online profiles.

The case studies involved 50 in-depth interviews with workers, including 21 ride-hailing drivers, 11 TaskRabbit workers, and 18 couriers. I drew upon platforms’ corporate discourses about ratings to supplement and contextualize the findings. The project was divided into two phases.² I began by studying ride-hailing platforms in the first phase of the research in 2017 and 2018. Most of the interviews with ride-hailing drivers (14 of 21) were conducted at that time. I set out, in 2020, to interview more ride-hailing drivers and include the other two cases for comparative analysis. Although I conducted interviews with gig workers during the early period of the COVID-19 pandemic, the primary goal of this study was to understand how algorithmic metrics shaped and intensified workers’ precarious experiences across labor platforms. Yet, particularly with delivery platforms, participants’ precarious experience and mitigation strategies became intersected with the evolving work environment during the pandemic.

As gig workers lack a shared workplace, I recruited interviewees online (e.g, posting recruitment messages on Twitter and worker-to-worker online communities) and through snowball sampling via interviewees’ networks. The interview sample varied in the lengths of platform-mediated work experience (from 2 months to 9 years), working hours (from 3 h to 72 h per week), and geographical locations (over 10 cities throughout the United States). The sample was skewed to male gig workers (40 of 50), partly because driving and delivery work are male-dominated industries and partly because female gig workers might be in a more vulnerable position to accept an interview request in worker-to-worker online communities.

Interviews were conducted over the phone and/or videoconferencing tools (e.g. Skype and Zoom) and were audio-recorded with participants’ permission. I used a similar interview guide in both research phases, while I refined the wordings to correspond to specific platform interfaces and features. Topics included participants’ background, uses of metrics and platforms, interactions with customers, acquisition of work-related information, and reflections on their work. Most of the interviews lasted between 30 min and 1 h. I followed a process outlined by LaRossa (2005), which is consistent with the grounded theory approach. I began with open coding, which identified the key terms and basic frames used by the participants. Axial coding identified the conditions through which the interviewees felt about metrics. Finally, the themes concerning the affective power of metrics emerged from selective coding. During the analysis, I paid attention to similarities and differences in participants’ responses to the affective discipline of platform-based metrics across the three cases.

In this article, I have examined how gig workers anticipated, interpreted, and managed algorithmic metrics across ride-hailing, delivery, and domestic services platforms in the United States. Specifically, most interviewees felt anxious about customer-sourced ratings because they are consequential to workers’ platform-based employability, whereas participants might feel ambivalent about system-behavioral measures depending on their varied considerations of the relevant disciplinary outcomes. Such differences illustrate that individuals can feel and experience metrics unevenly (Beer, 2016). The analysis finds the convergent reception of customer-sourced ratings because of the platforms’ unequal treatment of customers over workers and the inherent unpredictability of customers in service work. In the three cases, workers anticipated and mitigated the influences of customer-sourced ratings at various stages of the work process. Metrics do not only exercise affective power after individuals are being measured (Beer, 2016). Nonetheless, the strategies for managing affective measures and customers only allowed them to momentarily routinize service interactions and regain control of their work. Echoing Cameron and Rahman's (2022) observations, workers have the most autonomy in the pre-hiring process in which they can filter difficult customers and strive to manage customers’ expectations. As service interactions proceed, workers have limited ability to influence customers’ evaluations and dispute ratings, thus intensifying their precarious experiences. TaskRabbit workers felt more satisfied with their work than ride-hailing drivers and couriers (see Schor et al., 2020) because Taskers were likely to have repeated interactions with customers which allowed them to cultivate a long-term interpersonal relationship. Alongside the settings of platform-mediated worker–customer interactions, the (in)visibility of metrics and platform dependence shape workers’ practices toward metrics.

Algorithmic management rests upon a trilateral relationship between platforms, workers, and customers (Stark and Pais, 2021). The evaluation of workers is not just about the management of distributed workforces, but also about building trust with consumers in anonymous marketplaces. Ratings help to enroll workers and customers “in the practices of algorithmic management without managerial authority having been delegated to them” (Stark and Pais, 2021: 47) at a low cost. While platforms discourage workers from filtering customers and declining orders, platforms may anticipate and encourage workers’ management of customers. The inducement of anxiety might be better understood as a technique of algorithmic management for cultivating self-improvement and performance management (Beer, 2016; Espeland and Sauder, 2016).

Importantly, gig workers’ past ratings and strategies might not always be relevant to themselves because customers represent a key source of work-related uncertainty. Management by and of customers remain focal points of production in the three cases. Rethinking algorithmic management through the lens of the service triangle (Leidner, 1993) allows us to situate gig work into long-standing, unequal power relations between managers, workers, and customers. Using Berlant’s (2011) idea of “cruel optimism,” Jarrett (2022: 60) writes “the state of being attached to a potential or possible future” incites gig workers to consent to exploitation. Workers become attached to gig work primarily because of the promises of flexibility and autonomy. Workers’ coping strategies ostensibly allow them to manage their metrics to keep their jobs or even gain a competitive advantage. However, this hope can become “cruel” when workers realize “the object/scene of desire is itself an obstacle to fulfilling the very wants that bring people to it” (Berlant, 2011: 227). For instance, there are tensions between maintaining high system-generated behavioral measures and the flexibility of accepting valuable orders. Workers’ everyday agency is often disrupted by difficult customers, information asymmetries, unfair dispute policies, and algorithmic opacity. The lack of actionable information also raises questions about whether workers can truly be treated as independent contractors when they cannot make informed decisions about their work. Gig workers are placed in a vulnerable position in which they are required to rationalize and cope with unpredictability in their everyday lives.

What is at stake here is not merely whether workers’ performance should be evaluated, but the point that platforms put too much emphasis on ratings and make it difficult for workers to dispute unfair ratings. Metrics can be useful for building trust among total strangers and providing reference points about service quality. However, the discrepancy between intended purposes and actual uses of metrics is a common pitfall when metrics become dominant modes of evaluation (Espeland and Sauder, 2016). Scholars and policy makers have directed attention to how constant pressure for reputation management “put workers under considerable strain which creates a working environment that fuels stress and create risks for PPVH [i.e. psychosocial violence and harassment] where unpaid but necessary work creates unmanageable workloads” (Moore, 2018b: 6). The 2021 European Commission's (2021) proposed directive has suggested that ratings should be considered as management tools and suggested labor platforms to “evaluate the risks of automated monitoring and decision-making systems to the safety and health of platform workers” (p. 37). Consistent with the Fairwork Project's (2022) principle of fair management, possible solutions are to have platforms provide a meaningful explanation of ratings and allow workers to appeal against platforms’ decisions.

The affective power of metrics goes far beyond the gig economy, especially considering how metrics play an ever-important role in shaping access to material and symbolic resources in our everyday lives (Fourcade and Healy, 2017). While this research compares the affective power of metrics across three types of geographically tethered gig work, further research is needed to document the convergent and divergent metric power in different work contexts. As in the three cases, we can observe that not all metrics matter, and even if they do matter, they may have unequal impacts on workers. Indeed, Van Doorn et al. (2022) argue that migrants make up a sizable portion of the gig economy, partly because platforms offer vulnerable migrants with seemingly flexible employment opportunities. There is an urgent need for considering how affective measures can disproportionally affect marginalized groups of workers and unevenly distribute OSH risks in the ever-expanding culture of metrification.

The author would like to express gratitude to Lee Humphreys, Brooke Erin Duffy, Malte Ziewitz, Tarleton Gillespie, and members of “New Media & Society Working Group” at Cornell University for giving feedback on earlier aspects of this project. I would also like to thank Matthew Zook for his editorial guidance and the three anonymous reviewers for their invaluable feedback. I am also grateful to the interviewees who generously shared their time and insights for this study.