Working under management algorithms
A motorcycle spins in the parking lot of a McDonald’s in South London. The driver circles the building trying to centre the GPS signal on his smartphone. He turns a few times, waiting for the delivery app, to which he is connected, to correctly track his position. Unless the app matches him to the specific coordinates of the food joint, he will not be able to pick up the order that has been assigned to him. In his workday he sometimes comes up with strategies to make sense of the software algorithm that governs his labour, and tries to use these insights to steer it towards his advantage. Often, there is little he can do. The algorithm behind the delivery app gathers information constantly and adapts in real-time. Consequently, it is more effective at moulding his labour conditions than he is at understanding and bypassing it.
The app finally finds his bearings and the driver can turn off the engine. It is Friday night, peak time. A group of eight delivery drivers have parked their vehicles and are waiting for orders to be dispatched. The delivery crowd wear mismatched uniforms and bags with logos corresponding to different companies. There is one element that unites them: the smartphone always attached to their hand, connecting them to the gig economy platform. The drivers wait. Waiting time is not paid, although it usually takes several minutes for each order to be ready. All they can do is stand outside McDonald’s. Some take a moment to connect their smartphone to a battery bank, others use this time to call home, to another time zone. The rest hang out in the entrance of the building. A few already know each other, they’ve been assigned to pick up orders in the same restaurant before. They only meet their workmates when customers’ choices happen to align, since the app is designed to conceal who their colleagues are. During these moments of idle time, they share their guesses on how the platform works, and complain about the systematic decline of pay.
Gig economy companies define themselves as the technical platform through which a customer finds access to a commodity. Platforms rely on software algorithms to manage goods, services, and workers. An increasing number of jobs – especially in food delivery – are overseen and organised remotely by management algorithms. Riders are in a position of asymmetry to the management algorithm that governs their activity. Their experience is transformed into data to design and optimise the performance of the delivery app. Through their smartphone, they provide their real-time location and all the information for each task performed, which is then fed to an algorithm that aggregates it to organise labour. The algorithm can then decide how to assign orders, design routes, or even arrange a pricing system through machine learning. Through the extraction of the data that is generated by the workers, they are integrated as individuals into the algorithm. Not only their labour, but also the mapping of the way in which they perform this labour are the source of profit for delivery companies.
In algorithmic management, delivery workers are completely subsumed to the algorithm. They find themselves engaged in a process of constant improvement of production, under continuously evolving rules that are unknown to them. For the past year, I have been researching these labor conditions shaped by algorithmic management. I have started to think about and experiment with an alternative approach to the data generated by riders. Can the data that is currently extracted from workers be repurposed towards their empowerment rather than for the revenue of platforms? Can data be disaggregated from its economic purpose, to be reaggregated towards building on workers’ leverage, improving their working conditions and giving them agency in decision-making?
This approach arose from conversations I had with delivery drivers in South London over the last year. As a member of the International Workers of Great Britain (IWGB) union, I started meeting with a group of drivers facing different workplace issues from delivery platforms. They initially contacted the union due to a problem of inaccurate mapping in the app that determines their journeys, used by a delivery platform called Stuart. Riders delivering on motorbikes found that the internal system of directions gave them bicycle routes. These routes, established by the algorithm, led through parks and down one-way roads, which motorcycle riders cannot follow without the risk of fines. Detouring means longer distances for an order to be delivered. Because of this their income, which is based on the mileage projected to get the order from the restaurant to the client rather than the actual distance, decreases.
Working with a group of three riders, we set out to organise a campaign to demand that Stuart fix the mapping system. We gathered around forty drivers in a WhatsApp group, but only ever managed to bring seven to the meetings. Out of this initial setback, there emerged the idea to rethink the way political organisation is carried out under delivery platforms. In what ways could riders hack the imposition of conditions of labour generated by the algorithm, to intervene in its architecture?
Guessing and gaming the algorithm
Delivery riders are constantly experimenting with ways of reframing technology to improve their labour conditions. Algorithmic software architecture atomises workers because algorithms are designed to increase competition between riders through rating systems and individualisation. They also isolate them by concealing the identity of their workmates from one another. Despite this, the system ends up bringing workers and riders together for long periods of idle time while they wait for orders. The platform also operates on a smartphone, a device that can also be used for deviant communication.
Both the physical and digital can become mediums for collective rule guessing. Through online forums, WhatsApp groups or informal gatherings in restaurant parking lots, riders start to share their knowledge and understanding of the algorithm. In this exchange of ideas, a collaborative interpretation starts to take place. Individual hypotheses can become collective theses, giving workers some agency in shaping their labour conditions. Collective ‘rule discovery’ contributes towards building solidarity between workers and can be considered as an act of resistance: riders overcome the atomisation of the algorithm by collaborating with each other at its margins. With a better understanding of the inner workings of the algorithm, riders can begin to engage in practices to ‘game’ the algorithm.
Riders may tamper with the GPS system or work across multiple platforms in parallel, playing off the thresholds of detectability of each company in order to deliver multiple orders at the same time. In some cases, platforms try to attract riders to areas where there is more demand by offering ‘surge prices’. In response, workers sometimes collectively log off the platform to allow orders to pile up, artificially creating surge areas that they can then all work in for slightly increased pay. This can be identified as a collective strategy to sabotage the algorithm.
Working across platforms is discouraged by companies and can cause the termination of accounts. Nevertheless, some riders have learnt how much deviation – in distance or in time – can take place before the algorithm starts penalising them. Based on this principle, they will take different orders that go in the same direction across platforms. First, they will deliver the orders from platforms that have narrow thresholds of detectability, and they will later deliver those from the platforms that are a little more flexible. In this way, they can increase their income.
Riders also outwit digital surveillance using fake GPS, which allows them to artificially relocate themselves. When being late to their shift they might use a fake GPS system to avoid being penalised. Some also use it to relocate themselves quickly next to restaurants to get orders assigned after delivering in residential areas. However, this practice is contested among riders. The use of fake GPS means that a few riders accumulate more orders, taking a larger share of the work that would normally be distributed between all riders. As a strategy it might game the algorithm, but it also falls under the individualising and competitive logic that the platform generates.
Sabot: Towards a digital commons
While spending time with riders in the McDonald’s in Old Kent Road (an area of South London) and trying to organise for Stuart to fix the navigation system, we started to come up with our own strategies to pressure platforms into change. It was difficult to assemble with the riders: firstly, because the isolation created by the platform makes it hard for them to come together; secondly, because their days are quite unpredictable. They cannot foresee in which part of town the platform might offer more work so they are unsure of where they will be. And thirdly, in their world everything is mediated by a smartphone. An assembly is a physical encounter that is outside of the digital sphere. Additionally, most drivers I met in McDonald’s were migrant workers. Many have families abroad, so they are always on their smartphone: either to work or to connect with their loved ones. This made me consider whether it would be possible to develop a political tool that subverted the principles of extraction and control of delivery apps to build more agency for workers, through their smartphone.
The proposal that I made to the riders in Old Kent Road was to develop their own app to organise: Sabot. The goal was to experiment with technology as a tool for empowerment, instead of a means of production. In the 19th century, discontented workers engaged in sabotage to restore their agency, in a movement that became known as luddism. The origin of the word sabotage can be traced to the French word sabot, which means wooden shoe. Workers would throw their sabots into the looms that were reorganising their labour and eroding their power in the workplace. In the 21st century, workers could use their own smartphone to sabotage the algorithm that integrates them for profit, using it to increase their agency and leverage with delivery companies.
In Sabot the same location information that is extracted in platform apps is used collectively, in real-time. This app could allow the creation of a digital commons, a real-time collective repository that can inform decisions. A digital commons is a form of collective property that can push platforms to become more than machines of profit for shareholders. The collection of data can be aimed to the advantage of those that generate it. Rather than information on each rider’s activity being used to individualise performance and feed the optimisation of the platform’s algorithm, it can be used to build on workers’ leverage, to improve working conditions and riders’ capacity to organise.
To dispute the monopoly of data to big tech companies, Subject Access Requests1 are used to obtain data generated by riders retroactively. This data can be used in court to prove a relation of employment between rider and platform. The act of requesting acknowledges that data belongs to platforms, and it can only be demanded (or hacked). Furthermore, the format in which companies hand over data is often partial or illegible, a strategy to hinder its analysis.
A digital commons, on the other hand, disputes the epistemological presumption of transparency holding truth accountable. As journalist Mike Annany and scholar Kate Crawford have exposed, transparency includes an affective dimension where “the feeling of seeing something leads to a feeling of control over it”.2 Openness of a system is equated to the verification of this system; being able to observe a system seems to be enough to confirm its truth. What I argue here is that this is insufficient to access data. To contest the asymmetry between worker and platform, workers need to be able to use data to also intervene in the system they are part of.
Obtaining data retroactively also presents one drawback: it cannot be used to make real-time decisions. Critical finance scholar Michel Feher explains the importance of understanding the temporality at which financial capital operates to respond to it. Workers can go on strike on a one-off basis, voters go to the polls once every four years. Investment firms are not making decisions intermittently and punctually, but constantly and continuously. This constant ability to influence decisions gives them priority over what happens in all aspects of our lives. Similarly to financial capital, algorithms operate in continuous decision making patterns. Any response to the ways in which they inscribe the conditions of work requires acting at their pace, in a permanent and ubiquitous interruption of the platform’s self-valorisation through extraction.
An app designed towards giving delivery workers agency would contribute to the already on-going process of collective ‘rule discovery’. This time it would not only rely on guessing, but instead would be informed by the same data that is establishing the algorithm’s rules. For example, riders could decide in which part of town there seems to be less riders to catch more orders and see the way that influences their income. They could also learn how to cross-reference the data of their movements with the data on their income, to gain a better understanding of their hourly wage. This could help them to identify the moments in which the pricing system of the platform undergoes changes and react consequently. If drivers could have all the data about their activity and the activity of their colleagues in real-time, it might change the way they work and support efforts for effective organisation. It would also dispel with riders’ isolation. If riders could locate their colleagues, they could meet up more easily to organise for specific issues. This data would be visualised in a map, which would give riders a sense of the magnitude of their force in the city, something which is currently unknown to them.
The goal is not to destroy technology by throwing a sabot to the looms. The aim has to be to mould it into something new. To understand how it is formed is to be able to steer it towards other purposes, towards imagining a future where humans are more than just an appendage to the machine.
1 In Europe and the UK, individuals can request to access and receive a copy of their personal data, and other supplementary information.
2 Annany, Mike & Crawford, Kate (2021) “Seeing without Knowing: Limitations of the Transparency Ideal and Its Application to Algorithmic Accountability”.
Julia Nueno is a visual researcher and engineer working at the intersection between data, technology and communities.