Abstract
This qualitative case study adopts a labour process analysis to unpack the distinctive features of capital’s control regimes in the food-delivery segment of the Australian platform-economy and assesses labour agency in response to these. Drawing upon worker experiences with the Deliveroo and UberEATS platforms, it is shown how the labour process controls are multi-facetted and more than algorithmic management, with three distinct features standing out: the panoptic disposition of the technological infrastructure, the use of information asymmetries to constrain worker choice, and the obfuscated nature of their performance management systems. Combined with the workers’ precarious labour market positions and the Australian political-economic context, only limited, mainly individual, expressions of agency were found.
Keywords
Introduction
The rise of the digitally enabled ‘gig economy’ or ‘platform-economy’ has been propelled by advances in computing power and information communication technologies (ICT) (De Stefano, 2016; Forde et al., 2017; Gandini, 2018; Howcroft and Bergvall-Kåreborn, 2019; Kenney and Zysman, 2016; Peticca-Harris et al., 2018; Srnicek, 2017; Wood et al., 2018). It has generated new ultra-precarious and commodified digitally enabled forms of labour, which can be viewed as a continued trajectory of neoliberalism (Zwick, 2018), including the increasing financialisation (Lapavitsas, 2011), fissurisation (Weil, 2014) and precarity of work (Rubery et al., 2018; Standing, 2016) – with some characterising platform-work as digital-Taylorism (Cherry, 2016) and others as post-capitalist (Peticca-Harris et al., 2018).
Theoretically, platforms intermediate between workers and end-users, instantaneously matching supply and demand to create two-sided digital-marketplaces for services (De Stefano, 2016; Peticca-Harris et al., 2018; Srnicek, 2017). Different typologies are used to categorise the various work arrangements facilitated by platforms (e.g. De Stefano, 2016; Howcroft and Bergvall-Kåreborn, 2019). Howcroft and Bergvall-Kåreborn (2019) distinguish, for instance, between four types of platform-work based on who initiated the work (either the worker or end-user) and whether the work itself is undertaken on a paid or speculative basis. Asset-based platform-work is one of their typologies, which captures real-world services, frequently organised via mobile phone applications (‘apps’). A subsection within this category is ‘work-on-demand via app’ (hereafter ‘app-based’) services, with the ridesharing behemoth Uber exemplary of this classification (De Stefano, 2016). In ‘on demand’ services, workers interact in person with end-users in a platform-mediated labour process. Platforms, as part of their business models, frequently monetise the data they capture and generate, with some portraying them as ‘rent-seekers’ who realise profit by taking a proportion of the transactions from end-users to workers (Peticca-Harris et al., 2018). This characterisation underplays, however, the active role that app-based platforms play in standardising services and shaping labour processes.
Arm’s-length contractual relations with labour (Cherry, 2016), structured through triangular or non-standard engagements (Rubery et al., 2018), are also critical to platforms’ business models; shifting the economic risks of work onto workers who must provide some assets to participate in the production process and are commonly remunerated on a piece-rate basis (De Stefano, 2016). Consequently, workers are frequently excluded from employment protections, have few guarantees for ongoing work and are exposed to the higgling of markets (Stewart and Stanford, 2017).
Most scholarly attention on app-based platform-work focuses on the distinctive features of the work, such as the reliance on algorithms to manage workers (Kenney and Zysman, 2016; Rosenblat and Stark, 2016; Zwick, 2018), and how these forms of work should be regulated (Cherry, 2016; Stewart and Stanford, 2017). Less attention has been paid to how platforms seek to extract and control labour effort in the production process or workers’ experiences of this work. This article, therefore, explores labour–capital relations in the app-based food-delivery segment, focusing on the Australian operations of UberEATS and Deliveroo. There are no precise figures concerning the size of their Australian operations, but both platforms offer self-reported statistics: Uber indicates it has 80,000 ‘delivery-partners’ across its ridesharing and food-delivery services (Uber, 2018), while Deliveroo has 3000 self-employed delivery-riders (Deliveroo, 2018a). Drawing upon qualitative interviews with workers from these platforms, this article explores the following research questions: How are the workplace control regimes configured by platforms in the Australian food-delivery industry and what are the distinctive features of these regimes? In addition, how do workers experience and react to these regimes? A labour process theory (LPT) analysis is used to examine these questions (Thompson, 1990).
Gandini (2018) contends that LPT provides an important resource for evaluating the digital evolution of work. App-based food-delivery work represents a technological repackaging and contractual reclassification of traditional food-delivery services. This form of work organisation can be characterised as novel due to the innovations platforms deploy to fix the organising problems associated with the division of labour and the integration of effort (Puranam et al., 2014). Both platforms offer multi-restaurant food-delivery services which match and coordinate interactions between workers, restaurants and consumers via their digital eco-systems while retaining flat organisational structures (Peticca-Harris et al., 2018; Srnicek, 2017). Organisational boundaries are further blurred by the adopted contractual relations, while their work allocation processes are increasingly dehumanised – using autonomously operating computer systems. Concurrently, these platforms rely upon perennial managerial fixes to labour–capital problems associated with capitalist production (Edwards, 1990; Thompson, 1990), including the use of piece-rates to pass demand-related risks onto workers (Finkin, 2016). The independent contractor classification, in turn, allows platforms to avoid other employment-related costs (Stewart and Stanford, 2017). LPT, through its explicit focus on the point of production and labour–capital relations, can reveal the distinctive control features of these novel forms of work organisation.
The article makes three key contributions. First, by executing an empirically informed LPT analysis it validates Gandini’s (2018) assertion that LPT provides a valuable resource to evaluate production relations in digitally enabled forms of work organisation. Second, it advances LPT by operationalising the analysis in a way that is sensitive to the wider political-economic context – an under-realised strength of LPT (Burawoy, 1983; Edwards, 1990; Smith, 2015; Thompson and Smith, 2009). The findings highlight the interdependence between platforms’ rationale for specific configurations of control and the broader political-economic context in which they operate, a critical insight on platform-work extending beyond food-delivery services. Third, by focusing upon worker experiences, the findings reveal how in the Australian context workers cope with and respond to labour process controls in these new forms of work organisation. The identification of predominantly individual, rather than collective, expressions of agency, which is in contrast to other jurisdictions (Johnston and Land-Kazlauskas, 2018), highlights the need to consider workers’ labour market profile and the broader political-economic context.
Theoretical lens
LPT analysis allows for the critical evaluation of labour–capital relations, particularly questions of control, resistance and exploitation (Smith, 2015); recognising the structurally antagonistic nature of these relations resultant from the exploitative character of capitalism (Edwards, 1990; Jaros, 2010; Thompson, 1990). In capitalist production, due to the relative autonomy of the labour process (Edwards, 1990; Friedman, 1977; Hall, 2010; Thompson, 1990), there remains indeterminacy in how structural factors affect labour process outcomes. Relations at the coalface of production are the result of distinct logics that arise from the terms of engagement, historical relations and other contextual factors (Edwards, 1990). A key challenge for capital in the labour process is the activation of labour effort, including the need to simultaneously obscure and secure the extraction of surplus value (Burawoy, 1979). In capitalist production, where labour is ‘free’ to sell itself to the highest bidder, capital has to manage the ‘double indeterminacy’ of realising labour effort and managing the relative mobility of labour (Smith, 2006). LPT privileges production relations as the point of analysis, enabling researchers to assess labour–capital struggles over the extraction of labour effort, distribution of profits and need for reproduction (Jaros, 2010); placing workers and their experiences at the centre of the analysis (Smith, 2015; Thompson, 1990). Workers do not routinely accept or acquiesce to managerial directives and controls (Coe, 2013) and are able to express individual and collective agency in the form of a variety of coping mechanisms and the ability to resist and rework managerial control (Katz, 2004).
Gandini (2018: 9) contends that LPT is a valuable approach to unpack how production relations in the platform-economy are configured. This study draws upon several control dimensions developed in second wave LPT to assess how capital seeks to realise and control labour effort in the food-delivery segment of the platform-economy, including: direct, technical, bureaucratic, normative and computer control (Braverman, 1974; Callaghan and Thompson, 2001; Edwards, 1979; Elliott and Long, 2016; Frenkel et al., 1995; Thompson and van den Broek, 2010). There is a trend towards an increased hybridity of control (Thompson and Harley, 2008), whereby capital applies control in complementary, interlinking and blended ways (Callaghan and Thompson, 2001; Thompson and van den Broek, 2010). Computer control, for instance, is a hybridised control dimension capturing the directing, monitoring and evaluation of work by software programs that rely upon big data (Elliott and Long, 2016). Similarly, normative control cannot be treated as an entirely discrete dimension, since ‘[a]ll control practices have normative dimensions’ (Thompson and van den Broek, 2010: 6). The normative control initiatives deployed by platforms to transform workers’ values, identities and emotions can elicit worker cooperation, strengthen managerial prerogative (Frenkel et al., 1995; Thompson and van den Broek, 2010) and commodify emotions at work (James, 1989), and hence should be identified.
Capital, to reproduce itself, needs to continuously innovate (Thompson, 1990). LPT’s ability to examine the changing role of technology in production facilitates the assessment of evolving labour–capital relations (Hall, 2010). New technologies, including big data, computer processing power and mobile phone technology, aid capital’s revolutionisation of control systems (Elliott and Long, 2016; Gandini, 2018; Hall, 2010). It remains unclear, however, how such technologies impact on labour, as they can be used for a range of organisational objectives (Hall, 2010). The platform-economy can be viewed as a technologically driven transformation of work (Gandini, 2018; Wood et al., 2018), whereby platforms’ digital eco-systems play a critical role in harnessing and galvanising both physical and creative labour effort (Howcroft and Bergvall-Kåreborn, 2019). The use of algorithms to manage workers and structure the production process is frequently flagged as a distinctive feature of platform-work (e.g. Lee et al., 2015; Schildt, 2017; Wood et al., 2018). A growing body of literature refers to ‘algorithmic management’ and ‘algorithmic control’ (Kenney and Zysman, 2016; Lee et al., 2015; Rosenblat and Stark, 2016; Schildt, 2017), signifying control systems increasingly based on ‘deep learning’ (i.e. self-learning algorithms) replacing management as an agent of capital. In such systems, machines are given the responsibility to make and execute decisions affecting labour, limiting human involvement and oversight over the labour process (Schildt, 2017). Characterising entire control systems as ‘algocratic’ (Aneesh, 2009; Lee et al., 2015; Rosenblat and Stark, 2016), where any form of control or task allocation is ascribed to algorithmic activity, is problematic and may lead to technological determinism (Boreham et al., 2007) while underplaying continued managerial discretion and oversight.
A final strength of LPT is its ability to move beyond the ‘vacuum of the shop floor’ (Edwards, 1990: 131) and assess production relations within the broader societal structures in which they are embedded. Platforms frequently seek to circumvent or re-make the regulatory regimes to which they are subjected (De Stefano, 2016; Srnicek, 2017; Stewart and Stanford, 2017). In the Australian context, platforms commonly classify workers as independent contractors rather than employees, limiting the controls at management’s disposal since the ability to exercise direct control under subcontracting arrangements is constrained (Stewart and Stanford, 2017). Hence, a critical assessment of the political-economic context under which the work takes place helps to discern the rationale for specific configurations of control (Callaghan and Thompson, 2001; Smith, 2015; Thompson and Smith, 2009).
Research methods
A qualitative case study approach with an influential case design (Yin, 2013) was used to assess the workplace control regimes of two food-delivery platforms operating in Australia: Deliveroo and UberEATS. The sampling focused on workers who operated on at least one of these platforms in the 12 months preceding the interview. The authors conducted 41 semi-structured interviews and one focus group. A total of 58 workers were interviewed between January and June 2017. In addition, three informal meetings were held with representatives from one platform between April 2017 and December 2018.
The platforms were selected due to their market share, operational areas and prominence within the industry. Riders did not necessarily work exclusively for one platform, with 24% indicating that they engaged in so-called multi-apping, using platforms interchangeably. Forty-nine per cent worked solely for UberEATS and the remaining 27% worked exclusively for Deliveroo. Workers used different modes of transport to undertake the deliveries: 75% used bicycles, 20% scooters or motorbikes, and the other 5% cars. Most participants were male (53 of the 58 interviewees), and a considerable proportion were temporary migrants (47 of the 58) on working holiday programmes and international student visas, from countries including Brazil, China, Columbia, France, India, Malaysia and the UK.
A multi-tiered participant recruitment strategy based on purposeful and opportunistic sampling was employed (Creswell, 2014), including snowball sampling techniques (Biernacki and Waldorf, 1981). Initial contact was made through face-to-face street intercepts in Melbourne and Perth and online recruitment via social media. Interviews were performed face-to-face and via telephone. Due to the different sampling approaches different interview dynamics existed. Street intercepts, for example, took place while workers waited for their next deliveries; infrequently these interviews had to be cut short and reconvened at a later stage. Interviews arranged via social media, in contrast, were scheduled at the worker’s convenience. The length of interviews varied as a result, ranging between 5 and 82 minutes and averaging 33 minutes. Using semi-structured interview protocols (Creswell, 2014), workers were queried about the design of the delivery process, their experiences of the work, what kind of controls they were subjected to, the role of technology in the labour process and what level of discretion they had over their work. The platforms’ performance management systems and workers’ interactions with the organisations were also examined.
A thematic analysis was used to code and analyse the primary data (Fereday and Muir-Cochrane, 2006). A hybrid coding process was adopted, coding the interview data in Microsoft Excel across six a priori defined categories: ‘the labour process’, ‘payment structures’, ‘control strategies’, ‘workers’ perspectives’, ‘agency’ and ‘coercion and consent’. Control dimensions derived from the LPT literature were further used to systematically code the different workplace controls identified in the interview data. Other aspects of the labour process and worker experiences were coded freely. Transcripts were reviewed on a line-by-line basis, coding the primary data across all relevant categories. Subsequently, core (e.g. job allocation process) and sub-themes (e.g. worker theories on the allocation process) were identified within the broad categories, allowing the researchers to evaluate the various facets of the labour process and control regimes. All authors contributed to the first round of coding. To ensure inter-coder consistency (Auerbach and Silverstein, 2003), the first and second author jointly coded several transcripts until consensus on the used codes was reached, briefing the third coder before commencement. Throughout the coding process the research team held regular discussions to refine the codes and themes. A second round of coding by the first author was undertaken to cleanse the data and ensure inter-coder consistency (Auerbach and Silverstein, 2003).
Labour process analysis of platform-based food-delivery work
The analysis of the two platforms’ labour processes revealed that they, although very similar, were shaped by distinctive organisational factors. Central to the platforms’ control regimes were three distinctive features: the technological infrastructure; deliberately created information asymmetries constraining worker choice; and the obfuscated nature of the performance management systems. These resulted in highly standardised services with geographically dispersed workforces. Despite constraints on workers’ ability to influence the labour process, expression of agency was found.
The labour processes
Before workers could operate on either platform they must register an account, procure the necessary equipment (e.g. bicycle, mobile phone with data and food-delivery bag), and install the worker app. Figure 1 summarises the labour processes of both platforms, which were straightforward and similar. There are, however, three important organisational factors that produced different labour process dynamics. First, Deliveroo restricted the pool of active workers by having a selection process and waitlist for new entrants, controlling the supply of workers and providing basic safety and customer training (Interviews 7, 33 and 35). UberEATS had few restrictions, providing a direct pathway for new entrants into the sector. As a result, some Deliveroo workers expressed a sense of superiority over their UberEATS peers (Interview 11).
The labour process of platform-based food-delivery work.
Second, in its Australian operations, Deliveroo initially used a core workforce engaged on ‘shifts’. ‘Shift-riders’ received a number of ‘guaranteed’ deliveries, a proxy for an hourly rate of pay, for the duration of their shift (typically between 30 and 150 minutes). Core workers were supplemented with precarious peers who were paid fixed piece-rates. UberEATS provided no guarantees, remunerating all workers on a variable piece-rate, which included consideration of the distance travelled.
Finally, the platforms used different methods to ensure that enough riders were active in specific areas. Deliveroo designated workers to specific zones and required them to ‘check in’ within these zones before they could receive orders (Interviews 1, 7, 11, 12, 28 and 32). UberEATS, in contrast, allowed workers to operate wherever the service was offered and relied on ‘economic nudges’ (Rosenblat and Stark, 2016) in the form of surge-pricing (Gandini, 2018) called ‘boosts’ to entice workers to areas of high demand – relying on gamification-from-above (Woodcock and Johnson, 2018).
Workers revealed how the platforms’ use of the independent contractor classification and piece-rates for most workers, to an extent, reversed the labour effort indeterminacy which is traditionally associated with employment (Smith, 2006). The deployment of piece-rate systems was also unconventional, as workers had little control over output (cf. Burawoy, 1979). Owing to the scarcity of work and pre-set rates for deliveries, workers were unable to extract more from the time–effort bargain. Apart from shift workers, riders frequently complained about time lags between orders and expressed concerns about earnings (Interviews 10, 17, 24, 26 and 28–31). ‘You can never pick it. It’s really unpredictable’ (Interview 25). Compounding this was excessive, unpaid, waiting at restaurants. While the apps indicated when orders should be ready, riders complained about regular delays (Interviews 2, 4, 6, 12, 13, 15, 17, 20, 30–32 and 37). This led some to contend that earnings were not commensurate to time spent on orders (Interview 30), while others exhaled in anguish: [S]ometimes I can see the restaurant just start to prepare the food when they [sic] arrive … Sometimes it’s just five minutes because it’s fast food … I can wait, it’s not a problem, but sometimes more, 20, 35, 40 minutes and the restaurant is busy – oh, it’s a long time. (Interview 13)
Distinct modes of control in app-based food-delivery work
Despite organisational differences, both platforms deployed similar control strategies and mechanisms to realise labour effort. Three distinct features of the control regimes emerged from the interview data: the role of technological infrastructure, the deliberately created information asymmetries, and the obfuscated nature of the performance management systems – discussed below. Both platforms relied upon multiple control dimensions, including technical, bureaucratic, normative and computer control. These were used in tightly interwoven, complementary and reinforcing ways. Technical and computer controls shaped the pacing and direction of the work and supported automated performance management processes. In conjunction with bureaucratic controls, these engendered productive worker behaviours with minimal, direct, managerial input and oversight. Normative controls, in turn, sought to ingrain a customer-orientation amongst workers, pushing them to invoke emotional labour in customer interactions (Gandini, 2018). Contrasting traditional normative control attempts by management, platforms cultivated worker commitment to the work and customer rather than the organisation, encouraging workers to view themselves as entrepreneurs – legitimising the independent contractor status. The self-entrepreneurial mindset was further instilled through organisational cultures which emphasised worker autonomy and flexibility while providing limited organisational support (Interviews 9, 17, 19, 23, 26, 29 and 32). Although some workers rejected these normative conceptualisations (Interview 30, see below), a considerable cohort bought into the entrepreneurial framing: ‘The best part is you can be your own boss. You can be your own boss and you can work as you wish’ (Interview 29). Critical and distinctive to the platforms’ control regimes are, however, three multi-facetted control strategies and mechanisms through which capital seeks to activate labour effort of these ‘non-employees’.
The app as the point of production: The importance of technological infrastructure
The platforms’ technical infrastructure – part of the digital eco-systems – was at the core of the control regimes (Peticca-Harris et al., 2018). This supports Gandini’s (2018: 7) characterisation of the ‘apps’ as the point of production, with workers describing these as pivotal to the work and interactions with the platforms: Everything happens on the app. (Interview 36) I mean, you’re really just following what the app tells you, so you don’t have to think much [laughs]. You just do what you’re told by the app. (Interview 31)
The app interface of both platforms contained several simple technical controls, including the need to accept or reject orders within seconds: ‘you have 10 seconds so you’re not really looking at it. You’re like, accept’ (Interview 9). Failure to adhere to these had serious implications for workers – discussed below. The apps also contained other technical controls in the forms of workflow protocols, like ‘order checklists’. Small differences between apps existed. UberEATS’ contained, for instance, a second-delivery function, prompting workers to collect multiple deliveries from the same restaurant (Interviews 9, 15 and 25). It further introduced an auto-acceptance function in the interview period; requiring workers to actively reject rather than accept orders (Interview 29).
Sophisticated computer controls were critical to the pacing and flow of the delivery processes, monitoring, evaluation and performance management. Three types of data fed into the computer controls: geo-spatial GPS data generated by tracking workers’ phones, worker–app interactions (e.g. cancellation of orders), and ratings by consumers and restaurants. Platforms relied upon these data points for the task allocation process. Geo-spatial GPS data was used to match riders with restaurants and customers as well as monitor workers in ‘real time’ during the delivery process. While apps facilitated the navigation process, workers explained they had discretion over the routes they could take, allowing them to take shortcuts or safer roads (Interviews 6, 7, 12, 13, 28 and 29) – revealing a level of responsible autonomy (Friedman, 1977): ‘You can choose your own [route] … So mostly I would follow Google Maps and if I find something that is better than the maps, like flatter roads, I will go with that’ (Interview 24). Autonomy was, however, constrained since riders were aware they were monitored (Interviews 6, 7, 11, 13, 33 and 40). Workers described calls and texts from the platforms when they headed in the wrong direction or took longer than estimated to complete orders: If it’s a certain amount of time after a delivery, or you forgot to press the button or something, then they’ll call you. I don’t think there is somebody breathing down your neck. I do realise that sometimes these operations people do say, I can see that you’re X distance from the restaurant or something, so they can see you in real time. (Interview 32)
Another crucial data point which shaped the computer controls was individual performance metrics – collected and communicated via the apps (Interviews 2, 6, 9, 10, 12, 17, 22, 25–29, 32, 33 and 37). UberEATS used three key performance criteria. Acceptance ratings were based on the proportion of orders accepted or rejected upon receiving a delivery request. Cancellation ratings captured the number of orders cancelled after acceptance. A customer satisfaction rating was produced upon completion of a delivery, with customers rating riders’ performance by giving thumbs up or down. In contrast to the purported information transparency that driver and consumer ratings provide in ride-sharing (Peticca-Harris et al., 2018), in food delivery these were not used to alert consumers or workers about each other’s previous conduct, and instead acted as a performance management tool on the UberEATS platform. Workers criticised the subjective nature of these ratings and explained how customers conflated errors or delays by restaurants with their performance – being blamed for leaky containers, cold food, misplaced items and wrong addresses (Interviews 10, 18 and 27): Sometimes you don’t have nice experience in Uber, because they are not organised and when they give you two orders, sometimes it’s very hard distance … The second customer will be hungry … [and] say, oh it’s cold, but it’s not my fault. (Interview 10)
Deliveroo also closely monitored riders’ performance using ‘service delivery standards assessment’ criteria, including: the time to accept orders, travel time to restaurants, travel time to customers, time at the customers, unassigned orders (comparable to UberEATS’ acceptance rating) and cancellation of shifts. Deliveroo’s performance reports informed riders whether their conduct met or fell below stipulated service level criteria, including their acceptance rating and average speed. Interestingly, Deliveroo’s performance metrics did not involve customer ratings, suggesting that ‘management by customer’ is not a universal feature of the platform-economy (Gandini, 2018: 11). Nonetheless, the digital eco-systems of both platforms were intricately configured with the clear purpose of shaping worker conduct in the labour process.
Information asymmetries
Platforms have been hailed for allowing markets to operate more efficiently (Peticca-Harris et al., 2018; Srnicek, 2017). Underlying this free-marketopia sits a reality where workers were compelled into activating labour effort due to imperfect information. Both platforms withheld critical information from workers, including the exact details of deliveries. Platforms only provided the pick-up locations (the restaurants) and concealed consumers’ addresses: ‘Once you get the restaurant, you confirm that you have picked up the order, that’s when you get the address’ (Interview 2). Opaqueness concerning the task allocation processes further influenced workers’ behaviour (Interviews 6, 9, 12, 19, 26 and 27), with workers unsure how different factors, such as their proximity to restaurants and performance ratings, affected their ability to receive orders: ‘[I]t’s random how it picks, really. We don’t know how it works’ (Interview 9).
Deliberately manufactured information asymmetries represented a distinct control mechanism, as the decision to withhold information rests with management and materially influenced the labour process. This prevented these supposed independent contractors from making informed decisions about the orders they accepted, curtailing their ability to screen for the most profitable deliveries.
Obfuscated nature of the performance management systems
Platforms signalled to workers that if their performance fell below the stipulated criteria there would be adverse consequences. Platforms partly realised labour effort through workers’ incomplete understanding of these bureaucratic controls and how they operated or affected workers’ ability to receive ongoing work. The performance systems were ‘black boxes’ to workers, with many under the impression that accounts would be automatically deactivated if performance fell below certain thresholds (Interviews 6, 25–27, 31, 32 and 37): I’ve never cancelled any of them. I try to accept all of them because if you drop below 85% [the acceptance rate], then you’ll – I don’t know – something happens like you have to be [stood] down for a while. Sometimes the app will crash [while] accepting a job, so some of them can … [reject] but not from my doing. (Interview 31)
Workers also perceived preferential treatment being given to workers with higher ratings: ‘[f]rom what they told us the app has been fine-tuned to give more order[s] to more efficient riders’ (Interview 2). For example, UberEATS’ (2018) worker-support forum highlighted that: ‘certain promotions may have minimum acceptance rate or completion eligibility requirements’. Similarly, Deliveroo (2018b) on its Roo Community emphasised that: ‘priority booking access will be given to the riders who provide the most reliable service’. Workers further reported receiving bonuses when they reached the highest local averages (Interview 7) or achieved personal bests (Interview 9). Pushing a customer-oriented mindset, invoking workers’ emotional labour – what Gandini (2018: 11) identifies as ‘“gamified practices”’ and the creation of ‘techno-normative’ control.
While some workers ignored the metrics (Interviews 10, 19 and 28), many strongly believed that they mattered, particularly for the task allocation process. Most were, however, unsure how it affected the labour process (Interviews 1–3, 9, 12, 17, 22, 26, 28, 29 and 33): I’m not sure how the [system] works, but I think if you accept 100% of your orders, you’re essentially renewed – once you complete you’re renewed to the top of the list of waiting orders. You accept these crappy orders, but your priority [gets them on] the top, so once the order is finished then you’re given another order quickly afterwards. If you take yourself offline, or if you reject an order, or you don’t accept an order, then your acceptance rate goes down and you’re given less priority. As I say, I think that’s the way they’re trying to manage these – the distance things. My strategy has always been to just accept any single order that I’m given. (Interview 32)
Informal discussions with representatives from one platform confirmed the ‘carrot’ and ‘stick’ narrative around performance metrics. They explained, however, that in practice low ratings did not necessarily result in deactivation, as platforms require large enough pools of labour to meet customer demand, and thresholds could be adjusted on a local basis. These discussions underscored that while purportedly an algorithmic control process, the inputs for deactivation were shaped by managerial decision-making – a critical detail obscured from workers. Platforms’ efforts to conceal the inner-workings and impact of their performance management systems was thus primarily a bureaucratic control lever that sought to elicit particular behaviours.
Worker responses
The hybridised nature of the control regimes enabled the platforms to produce tightly managed labour processes. This did not, however, mean that workers simply accepted and acquiesced to the controls they encountered. Workers responded in a variety of ways, with most expressions of agency in the form of individual resilience and reworking (Katz, 2004). Nonetheless, some collective agency was expressed, including the sharing of resources and information as well as assisting others in case of accident or breakdown (Interviews 9, 11, 13, 17, 21, 23, 27, 29 and 33).
Individual expressions of agency were found, for example, in response to the platforms’ attempts to direct workers to work in particular areas (Interviews 1, 4, 15, 26, 27, 29, 36 and 38), at particular times (like New Year’s Eve) (Interview 9) or under certain conditions (e.g. inclement weather) (Interview 23). The platforms used different economic incentives, including boosts and bonuses, and normative messaging in the form of peppy text messages to induce workers to exhibit desired behaviours. While a large proportion of workers were interested in maximising their earnings (Interviews 1, 4, 6, 9, 26, 27, 29, 36 and 37), others argued that they ignored such attempts: ‘You know what? I don’t give any notice of them’ (Interview 41). Akin to ‘making out’ (Burawoy, 1979), some workers described how they restricted their labour effort (Interviews 7, 8 and 29) and only worked until self-imposed targets: ‘I just make up something in mind, like if I get at least $80 or $90 or $100 okay that’s enough’ (Interview 29). Although platforms portrayed themselves as funky, flexible organisations, their normative attempts to instil a customer-orientation and self-entrepreneurial spirit amongst workers were not universally successful, with some riders highly circumspect about these organisations’ cultures and intentions: So, they send a message; for example, Uber is very excited to announce the new boost and that new boost will be a lower one. They said, after listening to your comments and stuff, and that’s bullshit because they’re only reducing the pay. I think that culture is – for example – for me, as someone educated, it doesn’t work with me. (Interview 30)
Other forms of agency included attempts to rework the labour process. Workers espoused different theories about how the task allocation process could be influenced, including ‘where’ and ‘how’ to wait (e.g. different mobility patterns) and attaining high-performance ratings (e.g. highest average travelling speeds). One worker (Interview 6) created a heat map to predict ‘magical’ locations with the highest chances of receiving orders. Others kept records about the most profitable times and work areas (Interviews 13 and 31). Some simply acquiesced to the perceived randomness of the allocation process: ‘[m]y philosophy is just stay in the one area. If it’s good it’s good, if it’s not it’s not’ (Interview 5). Workers further explained how they reframed their delivery work into a video game (Interviews 6 and 9) or exercise routine (Interviews 7, 9, 13, 20, 25 and 36): I really enjoyed last Saturday. It was a nice day and I was just riding my bike. It feels like a video game. When you first started doing it – like Taxi. That’s exactly what it felt like. It’s fun. (Interview 9) I love cycling and it’s getting outside and the exercise – you know – but I don’t do this job for the money. I do it for the enjoyment more than anything. (Interview 25)
Blurring of work and play can be viewed as gamification-from-above since platforms relied upon a plethora of psychological tricks (Rosenblat and Stark, 2016; Scheiber, 2017) to influence workers’ normative perceptions of the work, framing it around ‘freedom’ and ‘fun’. It can, however, also be constructed as a coping response to alienation of their labour (Interviews 1 and 41) and hence a form of gamification-from-below (Woodcock and Johnson, 2018).
More extreme expressions of agency were the manipulation of geo-spatial data, theft of food (Interviews 15 and 41) and evasion of shift work (Interview 7). To circumvent computer controls, one worker manipulated his GPS signal through a location-masking tool, while further suggesting he hacked and deactivated the app’s auto-acceptance function, thereby providing himself with unlimited time to accept orders (Interview 9). Such detailed reworkings, however, were exceptional. Collective expressions of agency within the Australian food-delivery sector were also largely absent, with limited collective action organised via pre-existing labour networks and unions (Smiley, 2018) with none of the interviewees indicting that they participated in these collective forms of protest.
Discussion
This empirically informed LPT analysis (Gandini, 2018; Thompson and Smith, 2009) makes several contributions to the growing literature on the platform-economy. Having initially established the novelty in work organising (Puranam et al., 2014) by app-based food-delivery platforms, it explores labour–capital relations within the Australian operations of UberEATS and Deliveroo. It shows how control regimes in this segment of platform-economy are hybridised and are more than algorithmic management, with three distinctive features pivotal to activating labour effort. First, the technological infrastructure of the platforms, based on technical and computer controls, is critical to the realisation of the labour process, with the ‘apps’ functioning as the point of production. Sophisticated computer controls enable platforms to monitor workers in real time. It is shown, however, that worker awareness of their mere existence already acts as a panoptic feature of the control regime, resulting in the activation of labour effort – akin to control regimes in call centres (Callaghan and Thompson, 2001). Second, platforms create information asymmetries to constrain worker choice, curtail their ability to make informed decisions and express agency. For instance, platforms withhold delivery addresses from workers when orders are offered, thereby ensuring that these are accepted even when uneconomical for these supposedly independent contractors. Third, the obfuscated nature of the performance management systems limits workers’ understanding of the bureaucratic controls, acting as another control lever and eliciting compliance with the work rules. Both organisations perpetuate the narrative that performance management systems are automated, and that decisions around preferential treatment and continued platform access are executed by algorithms. Management’s role in calibrating and intervening in these systems appears deliberately underplayed. Combined, these distinctive control features enable platforms to produce tightly managed labour processes with marginally attached workers who are geographically dispersed.
The findings underline the need to consider the nexus between specific labour process control regimes and the political-economic context. In the Australian context, two structural factors are critical to understanding the rationale for the specific configurations of the control regimes in the app-based food-delivery segment of the platform-economy. First, these platforms seek to shift the economic risks of work onto workers and reduce their labour-related costs by classifying workers as independent contractors. This, however, limits the extent to which they can direct and control the labour process (Stewart and Stanford, 2017). By hybridising and obfuscating controls, platforms are able to mask the extent of their control over the labour process; enabling them to deliver highly standardised services with ultra-precarious workers, while at the same time publicly advancing the appropriateness of their continued reliance upon arm’s-length relations with labour – imperative to their business models. This platform-capital approach extends beyond food-delivery services (see e.g. Peticca-Harris et al., 2018). Second, food-delivery workers are paid per delivery and can engage in multi-apping. In order for food-delivery platforms to offer their services to consumers, similar to other capitalist organisations, they need to solve the double indeterminacy of labour power (Smith, 2006). What is significant about this segment of the platform-economy is how the assessed platforms are able to reverse the labour effort indeterminacy to an extent, placing the onus on workers to maximise the time–effort bargain. Platforms are, furthermore, willing to trade the fidelity of workers, to ensure that their obligations to workers and the risks of employment are minimised. App-based platforms, therefore, come up with novel labour process control fixes to the labour–capital indeterminacy while seeking to minimise their regulatory risks.
The political-economic context is further important to the realisation of labour effort. While workers enjoy relative mobility and flexibility, they have little control over how many deliveries they receive. Even in a developed market economy such as Australia with socially protected employment, platforms are able to carve out pockets of market despotism (Burawoy, 1983). This phenomenon most clearly plays out on the UberEATS platform, with its few entry barriers. Workers described increasing scarcity of work (Interview 30) and growing inter-worker competition (Interviews 1 and 6). These despotic market conditions act as another mechanism by which UberEATS ensures workers activate their labour power and comply with the work rules.
Finally, with platform-work a new stage in capital’s continued re-organisation of the production process, it is also essential to consider worker experiences of the labour process and their responses to the control regimes. The findings reveal that agency is expressed in a variety of ways, but mostly remains at the level of individual resilience and reworking (Katz, 2004), with collective expressions largely absent. This is in contrast to other jurisdictions, including the United States, United Kingdom and elsewhere in Continental Europe (Johnston and Land-Kazlauskas, 2018). The lack of collective action in Australia can be explained by two antecedents: the workers’ labour market profile and the Australian political-economic context. The migrant and transient status of many app-based food-delivery workers limits their ability and willingness to express collective agency (Coe, 2013). Migrant workers are at higher risk of exploitation (Anderson, 2010), while transient workers are willing to tolerate poorer working conditions (Galván, 2012). The political-economic context is further significant since their migrant status precludes these workers from social security provisions, increasing the need to earn an income through paid work. Some interviewees used platform-work to evade restrictions on their visas (Goods et al., 2019). Despite low levels of unemployment, Australia has historically high levels of underemployment (Oliver and Yu, 2018), suggesting there is a considerable reserve army of labour who can undertake this work. Coercive market pressures and the presence of vulnerable workers allow platform-capital to extract surplus value on the basis of a low-road labour costs model in a developed economy with minimum labour standards.
While the findings help to understand the nature of workplace control regimes in the app-based food-delivery sector, there are several limitations to the study. First, only one part of the platform-economy within one liberal market economy is assessed here, raising questions about the reflectiveness of these control regimes for the wider platform-economy. It reveals the need for more comparative research which encompasses other jurisdictions and forms of platform-work. Second, the article draws primarily upon worker experiences, since only limited, informal access to platforms was enjoyed, therefore future research on management’s role in the labour process of different forms of platform-capital remains critical.
Footnotes
Acknowledgements
This article would not have been possible without the valuable support of many colleagues. We thank in particular Ray Fells for encouraging us to use labour process theory, Trish Todd for her support in the early stages of the project, Arne Kalleberg for his feedback on our early work, Diane van den Broek for her critical insights on technology and work, and Bradon Ellem and Marian Baird for their ongoing support of the project. Lastly, we would like to express our gratitude to Dr Ian Roper and the three reviewers for their guidance and helpful feedback. Needless to say, any errors of fact or interpretation are those of the authors.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
