The Rise of Algorithmic Management

Algorithmic management is becoming an increasingly normalized part of industrial relations. A recent OECD survey found that 90 percent of workplaces in the United States use at least one form of algorithmic management (AM). That figure is only slightly smaller in Europe, where the AM adoption rate is 79 percent. In the US, three out of four firms used ten of the fifteen AM tools identified in the survey, while the intensity of AM tools was lower in Europe, with most firms using three to five tools.

What algorithmic management does for companies is compress space and time when it comes to the information they receive and their ability to make decisions based on that information. Before digitized algorithms, companies had middle managers to note and supply feedback on worker performance, customer demand, and so forth. Based on that feedback, executives would then give new instructions to optimize labor productivity.

With algorithms, that information flow is condensed into a set of automated and semiautomated processes. Almost instantaneously, data is crunched and new instructions are given out, generating a constant loop of data calculations.

This data revolution is a form of industrial relations power for bosses over workers. Executives can see, in real time, hundreds of data points on their workers, from how fast they are working to how many times they click on and off their work app. Meanwhile workers often have little more access to information on their own work than they had predigitalization. This gaping information asymmetry gives companies the power to use workers’ own data against them.

In the case of intensively surveilled Amazon warehouse workers, we can see how management can exploit this information asymmetry. Workers are tracked through biometric data, scanners, wristbands, and surveillance cameras. They can be punished or even fired if they perform poorly on metrics like tasks completed per hour, or if they are spending more time at the toilet than colleagues. In the dark about their data and that of their colleagues, under constant pressure to work fast, these atomized and exhausted workers suffer higher rates of injury and spend more time off sick than industry norms.

Another example is an Uber driver subject to the Orwellian-named “upfront fares” policy, where pay rates are algorithmically determined, based on a “black box” of data inputs unbeknownst to the worker. Since Uber has the data history of each driver, they can personalize pay rates so that they pay only as much as they think the individual driver will accept based on their history of acceptance rates and no more. This affront to the principle of equal pay for equal work has been used to deliver pay cuts by stealth.

While Amazon and Uber may be two firms at the cutting edge of exploitative algorithmic management practices, many companies even in unionized workplaces are introducing simple forms of algorithmic management, often using off-the-shelf systems, that are steadily eroding the autonomy of workers and tightening the control exercised by bosses. A recent survey of trade union members in Europe found one-third were aware of algorithmic management “used in recruitment, surveillance and daily decision-making of workers’ lives.”

There’s no doubt that bosses have reaped the rewards of the data revolution so far, but there is no reason why that must continue to be the case. For workers to rebalance the scales of industrial relations in the digital age, it is increasingly imperative that they come to grips with data.

Negotiating the Algorithm

Labor law professor Valerio De Stefano coined the term “negotiating the algorithm” to describe collective agreements between employers and workers’ organizations over “the use of digital technology, data collection, and algorithms that direct and discipline the workforce.” There are now various examples of collective agreements that negotiate the algorithm, although they remain the exception rather than the norm.

IBM’s German subsidiary has an agreement with the works council on the use of AI systems in the workplace that requires that AI must be transparent, including the ways in which input data influences the AI’s decision-making. All AI decisions must be traceable, so that workers can understand exactly how an algorithmic decision was made, and there must be identifiable human beings who can be held accountable for what the AI does. Workers also sit on an “AI Ethics Council” that evaluates risks from the AI, including risks to labor rights.

In the platform economy, the United Federation of Danish Workers (3F) union in Denmark has a collective agreement with home-cleaning platform Hilfr that includes provisions on algorithmic management. For example, Hilfr must supply a comprehensive explanation for all algorithmic decisions; if it cannot justify an algorithmic decision, it is deemed invalid.

Employees have codetermination over health and safety issues, meaning they can override any algorithmic decision that puts their health at risk. There is also a “digital clubhouse” on the Hilfr app for workers to engage with the union in various ways, such as the holding of ballots and the election of union representatives.

In Spain, the Just Eat food delivery platform has a collective agreement with the Workers’ Commissions (CCOO) and the General Union of Workers (UGT) unions that contains a chapter on algorithmic management. This includes the right to information on the “parameters” of Just Eat’s AI system, including “the rules and instructions that feed the algorithms.” An “Algorithm Commission” made up of employer and union representatives provides oversight to ensure the provisions in the chapter are being applied.

Collective agreements that control and limit algorithmic management should be the aim for all unions. However, in a hostile industrial relations environment where many companies are refusing to recognize unions and even establishing yellow unions as a way to sabotage worker organizing, unions cannot always get to the negotiating table straight away. In this context, unions need to be creative in how they recover, analyze, and make use of data as a means to build their power.

The use of “adversarial” data tools to recover workers’ data is an increasingly important organizing method in the age of AI, for three reasons. First, data can be used to inform union strategy. By knowing what the boss knows, unions can understand the company’s weak points and the reality of working conditions, helping them make better decisions.

Second, even if your union has a collective agreement that includes ostensibly strong data protections, how can you be sure the company is complying with that agreement without data tools to monitor, test, and verify that they are compliant? What data is being gathered on workers and how it is used may not be apparent to workers themselves; unions need data methods as a compliance measure.

Third, data is evidence that can back up the union’s argument and apply pressure on the company. If the union can recover data that proves workers are on average being paid below the minimum wage, for example, this can be the basis for a union recruitment drive, a media campaign, political lobbying, or even a lawsuit.

There are various case studies internationally that prove the potential value of data tools to worker organizing, although most of them up to this point have been in the platform economy. In Switzerland, Uber drivers worked with data scientists to piece together how much the company owed them, after a labor court found that they were employees and were owed backdated pay from 2017 to 2022. The data scientists found that the drivers were owed 20 percent more on average than what Uber’s calculations claimed.

In Brazil, hundreds of thousands of rideshare drivers have used an app called “StopClub.” It takes trip offers from platforms and instantly works out what the earnings per hour and per kilometer will be, so that drivers can easily decide if accepting or rejecting a trip offer will be economical for them. “It’s as if a blindfold has been removed from our eyes,” one driver said of StopClub.

In the United States, Armin Samii, an UberEats rider, developed an app for riders called “UberCheats” that used GPS coordinates to check if riders had traveled the distance that UberEats had claimed. Out of 6,000 trips logged by riders using UberCheats all over the world, the app found 17 percent to have been underpaid by an average of 2.2 kilometers per trip.

UberCheats is an example of what journalist Cory Doctorow has called “counter-apps” — apps that are specifically developed to provide easy access to data that can enable worker resistance. Or, as Doctorow puts it, helping “workers seize the means of computation from their bosses.” The experiments so far in counter-apps are just the tip of the iceberg, but the fact that almost all of them have been grassroots, DIY efforts shows that it is possible to use effective data recovery tools and techniques at low cost and with limited expertise.

Moreover, counter-apps are at the more complicated end of the range of potential data recovery tactics. In states where workers have data rights, making data requests to companies for information can be a frustrating process, but it nonetheless can yield results.

Data can be scraped from what the platform shows you on the app. Workers can also conduct controlled tests on the platform to find information, as many rideshare drivers have done to prove that they are subject to different pay rates for the same work. An app can be reverse engineered to find out what data the company is collecting.

Naturally, all of these methods have their strengths and weaknesses. Workers are likely to get the best results by using them in combination with one another.

Building Union Capacity

Of the many pioneering efforts in worker data recovery, remarkably few have emerged from within formal union structures. A study by the International Trade Union Confederation found that over 60 percent of trade union activity on algorithmic management consisted of analysis, awareness raising, and developing strategies/principles. Twelve percent concerned union organizing and campaigning, while just under 10 percent involved training and capacity building. “There is an urgent need to move from principles and theoretical discussions to implementing these principles in practice,” the ITUC report concluded.

In a sense, it should not be surprising that innovation has mainly come from the fringes of the labor movement when it comes to data. These tools are in their infancy, and unions have tried-and-true methods built up over decades (or even centuries). Furthermore, most reps and organizers are already rushed off their feet. Who has the time to add another layer of work that requires training and capacity-building to master and is not guaranteed to deliver results?

This attitude may be understandable. However, unless it is overcome, it is likely to hold unions back in organizing the working class for the digital age. Unions that are not data-savvy may struggle to appeal to a new generation of workers who have only known smartphones and apps. As AI systems are integrated into standard work settings, negotiating the algorithm will no longer be a fringe concern.

While the most pressing need for worker data recovery still lies in the platform economy and factories organized along digital Taylorist lines, these are merely the jobs at the cutting edge of the “datafication” of the economy. Even where the use of automated monitoring and decision-making is currently limited to a scheduling tool, or to tracking when employees enter or leave the office, these systems can still have significant effects on how workers are managed. Having a union that pushes management on data and algorithmic management can act as a disincentive to managers to introduce more intensive AI systems.

As a minimum, unions should be aiming for all reps and organizers to have undergone training, so that a basic level of knowledge and skills in “negotiating the algorithm” can be established. More ambitiously, unions should be seeking to build an in-house data team which can work with reps and organizers to develop a data strategy for each firm/industry. This should include providing technical support for data recovery and developing sector-specific policies for what collective bargaining over data should look like.

In Europe, the urgency to build union capacity has increased with the passing of the EU Platform Work Directive, which has to become law in all twenty-seven member states by the end of 2026. The Platform Work Directive establishes a broad set of rights for platform workers in relation to algorithmic management, including the right to an explanation and human review of automated decisions and the right for worker representatives to be consulted before substantial changes are introduced to AI systems. Platform workers will need to be part of unions where an ecosystem of data knowledge and skills is present to make effective use of these new algorithmic rights.

Workers have been gathering information about their work to aid their organizing efforts for as long as unions have existed. The only difference today is that, in the data age, there is a lot more information that can potentially be gathered and a variety of tools are required to gather it. The goal, as one worker told us in the research for our European Trade Union Confederation study, should be that “whatever the bosses can see, we can see as well.” Tackling capital’s information dominance over labor may seem daunting, but it is both possible and necessary.