Prague, Czech Republic – Trade unions are taking on algorithmic management in Amazon warehouses in Central Europe, with some success.
Like many residents of Most, a former mining town near Czechia’s northwestern border with Germany, Michaela Makova was in the red.
For 25 years, Makova had worked a string of low-paid jobs – from bakery worker to post office clerk – and the debts of a single mum had built up.
One in five people in Most faces foreclosure; at 7.5%, unemployment is twice the national average, the result of the gradual collapse of heavy industry, particularly the lignite mining around which modern Most was built in the 1970s.
So when Amazon set up shop on the outskirts of the capital, Prague, in 2015, many Most residents rushed to apply; the hourly pay rate of 6.5 euros, Amazon’s willingness to overlook bad credit when hiring and the offer of free transport to and from work was compensation enough for the 160-kilometre round trip each day. Makova, 53, joined in 2017.
Three years later, in August 2020, Amazon terminated her contract, citing loss of medical fitness. Máková had begun experiencing pain, numbness and tingling in her hands, the result, she says, of repetitive motions over an extended period that involved packing 2,000 delivery packages in a single 10-hour shift.
The pace at which she worked was monitored by a computer telling her exactly where to find each product, part of the algorithmic management that has become standard in warehouses run by the US tech giant as it races to satisfy demand from online shoppers.
In April 2021, Makova had surgery on her right arm to release nerve pressure from carpal tunnel syndrome. Two months later, the Hygiene Authority of Czechia’s Central Bohemian Region ruled that Amazon was responsible for her injury, giving her the right to claim compensation.
Amazon did not comment directly on Makova’s case, but an official, who spoke on condition of anonymity, told BIRN: “Safety is integral to everything we do at Amazon, every day, in every operation.”
Makova, however, recalled pressure to perform and confusion over how Amazon’s ‘algorithmic management’ evaluated her.
“It’s a pressure, having to assemble 500 orders an hour or pack 150 to 250 boxes,” she told BIRN. “We didn’t understand how the evaluation system worked; it was based on some kind of an average. Several times a day, we would get an evaluation where the manager would tell us we should work faster.”
It’s the kind of pressure that’s becoming the norm.
Logistics centres are booming in Central and Eastern Europe, currently occupying almost 40 million square meters of land in Czechia, Poland, Slovakia, Romania and Hungary. They symbolise, according to Daniel Sitera, a researcher at Prague’s Institute for International Relations, “the shift of capitalism towards the digital economy”.
With the end of the Cold War three decades ago, the region became a magnet for major Western manufacturers lured by its location, an abundance of cheap workers and less restrictive labour rules than in Western Europe.
The tech giants of the digital economy are no different, said Sitera. And their arrival has brought with it Artificial Intelligence and algorithmic management, with consequences for the physical and psychological health of employees that are only just beginning to be understood.
Trade unions, however, are fighting back. And they have landed a few punches.
BOX 1: Rule by algorithm
The term algorithmic management describes a type of management where workers are directed, evaluated, and disciplined by algorithms – rather than by their bosses – based on data previously collected about their behaviour.
According to the study ‘Algorithms at work: The new contested terrain of control’, published in 2020 by Academy of Management Annals, algorithmic control at the workplace is comprehensive to a degree that is unprecedented, allowing managers to know at all times what workers are doing, where and how fast.
But it can lead to frustration, insecurity and alienation of workers, who often do not understand what they are doing and why and become completely dependent on their computers, phones or scanners.
Algorithmic management is described by the authors as a direct successor of technical management whereby workers are ‘locked in’ certain positions in the work process and dictated their work rate by machinery (e. g. assembly lines).
While algorithmic management is increasingly popular in different types of employment [e.g. call centres, food delivery, taxis services], Amazon’s network of warehouses is an example of how algorithmic management can be used in logistics.
“Nobody tells you exactly how it works”
The vast majority of Czech and Polish Amazon workers – ‘associates’ in the company jargon – get their work orders either from a scanner or a computer, speeding up procedures in sprawling warehouses that store a vast range of products.
Every item and every shelf has a unique code, which workers scan. The act of scanning an item theoretically makes it possible for Amazon to trace not only the movement of goods, but of its employees as well.
Besides guiding workers through warehouses, the software also has an evaluation and feedback function embodied in Amazon’s ‘proprietary productivity metric’ known as ADAPT, or Associate Development and Performance Tracker.
In practice, this means that employees are evaluated weekly by ADAPT, and if they fall short of what is required, they receive a warning and coaching. After three such warnings, Amazon either requires them to sign a mutual termination agreement or can unilaterally dismiss them, according to workers, former workers and court documents consulted by BIRN.
Little is publicly known about how ADAPT quotas are calculated.
According to internal company regulations from 2015 submitted to the district court in the western Polish city of Poznan in 2016, each worker is evaluated in terms of attendance, quality of work and the achievement of minimum productivity rates based on the results of 90% of employees over the previous four weeks.
Makova is far from alone in struggling to understand how it all worked and therefore what was expected of her.
Another Czech Amazon employee told BIRN: “Nobody tells you exactly how it works, which is frustrating. Sometimes you think you are doing well and then you find out you are not.”
“Managers and instructors give you tips on how to improve, but these often don’t work. I realised that even they are just guessing how to make the system work for you.”
How an employee is navigated around the warehouse and how much they depend on the scanner varies between positions. Those who store incoming goods place them onto shelves wherever there is space.
Pickers are told by the scanners exactly where they should go and how long it should take them to retrieve an item, racing against a countdown in the corner of their screens. Those who assemble and pack goods, like Makova, stand in one place and receive instructions from a computer; if they need a toilet break, they have to find a replacement.
Some workers, like those loading trucks, are not subject to ADAPT. But they have to load all the goods sent their way by their ADAPT-led co-workers.
Amazon work contracts in Czechia and Poland – seen by BIRN – do not stipulate a worker’s precise position, allowing the company to switch workers around depending on needs. Some of those BIRN spoke to said this method of juggling workers can be abused.
“If the manager doesn’t like you, he puts you in a position where you are unable to meet the quotas, the system gives you three warnings and then automatically fires you,” said Ivo Mayer, president of Dobrovíz warehouse trade union at the Amazon logistics hub near Prague.
But it can also be advantageous. “When you have good relations with your manager, he may put you in a different department which doesn’t look so much at the quotas,” said a former team leader in Dobrovíz who declined to be named.
Asked about the ADAPT system, Amazon told BIRN:
“Like most companies, we have a system at Amazon that recognises great performance and also encourages coaching to help employees improve if they are not meeting their performance goals.”
“Performance metrics are regularly evaluated and built on benchmarks based on actual attainable employee performance history. We look at the performance that associates are naturally setting and then set the benchmarks from there with a focus on safety in mind.”
“Secondly, associate performance is measured and evaluated over a long period of time as we know that a variety of things could impact the ability to meet expectations in any given day or hour—this means associates would not be coached for having a single tough day.”
Asked specifically about complaints that the juggling of workers can be abused, Amazon said its objective was “to hire and retain talent – not the opposite.”
“We support people who do not perform to the levels expected of them, offering them dedicated coaching to help them improve.”
BOX 2: Expansion into CEE
Amazon opened its first and only Czech fulfillment centre in Dobrovíz near Prague in 2015, hiring 3,000 permanent workers.
The bulwark of Amazon’s expansion to Central and Eastern Europe is neighbouring Poland, where Amazon currently employs 18,000 people in nine warehouses, runs a development centre in Gdansk and opened a local e-commerce site in early 2021. Both the Czech and Polish Amazon warehouses were established primarily to service the German market.
Question mark over legality of data processing
So far it has been the power of trade unions and level of privacy and labour regulation that has determined the differences in Amazon productivity and monitoring practices in different countries.
Trade union lawyer Michal Sobol said that under Polish labour law, unilateral terminations require genuine, reasonable grounds and cannot be tied to specific productivity results.
In Czechia, the reasons for which a company can dismiss an employee are explicitly stipulated in the law. Because there is no specific reference to ‘productivity’, Amazon can cite “unsatisfactory work results”, “violation of discipline” or, as in the case of Makova, “losing medical fitness” in its justifications for unilateral contract termination.
If an employee with three ADAPT warnings agrees to sign a bilateral termination agreement, Amazon does not need to provide any justification. Those hired via agencies or on temporary contracts can also be dismissed without any necessary justification.
On the European level, the most powerful legal tool available to rein in algorithmic management is currently the General Data Protection Regulation, GDPR, which stipulates that collecting and processing of personal data must have a lawful basis, established either by consent of the data subject or by proof of a legitimate interest behind such processing.
Amazon says that it only collects data from the scanners to monitor movement of goods and to ensure worker productivity. “We definitely do not monitor the movement of workers,” said Miroslava Jozova, Amazon’s spokeswoman for Central and Eastern Europe.
The company, however, knows not only where individual items are located, but also who put them there and when. According to the former team leader who spoke on condition of anonymity, those in his former position have the capability to trace the movement and productivity of individual workers both in real time and retrospectively. Other Amazon employees BIRN talked to spoke of feeling like they were under surveillance.
To comply with GDPR requirements, Amazon requires all new hires to sign a privacy notice. But this is non-negotiable, and the Czech data protection authority, ÚOOÚ, has questioned whether it provides sufficient legal basis for data processing.
“According to the European Data Protection Board, it is improbable that an employee could freely reject data processing requirements of its employers,” ÚOOÚ told BIRN. “Employee consent therefore should not be used as the legal basis for data processing in contracts.”
Neither ÚOOÚ nor the Czech Labour Inspectorate has taken a deeper look at the legality of data processing in Czech Amazon as no official complaint has as yet been filed. In Poland, a GDPR complaint filed by the trade union Inicjatywa Pracownicza, which claims Amazon is illegally profiling its workers, is currently before the Warsaw data protection authority.
Workers in “de facto race” with each other
The efforts of Inicjatywa Pracownicza (IP) illustrate the pivotal role trade unions can play in reining in companies like Amazon.
Last year, the Polish trade union won a case on behalf of its member and a former Amazon employee, Maciej Gorajski, who joined the company in Sady near Poznan in 2014 but was dismissed in 2016. Amazon, according to court documents, blamed low productivity results. IP went to court, arguing that the dismissal violated Polish labour law.
In 2019, the district court in Poznan came down in favour of Gorajski, ruling that the ADAPT system was “very unfavourable to the employee” and therefore, “termination of the employment relationship must be considered unlawful”.
“This system creates a de-facto race between the workers who want to avoid having below average results,” the court ruled. It said that the minimum production rate set by Amazon “is not a constant known to the employee and it does not guarantee him certainty of the correct performance needed to maintain employment.”
The court also said that Amazon’s evaluation system failed to take into account when workers achieved targets above the minimum quota. The first instance decision was confirmed by the regional court in Poznan in 2020.
Gorajski never returned to Amazon, but his case was an important victory for IP. According to the trade union, in Poland, Amazon workers on permanent contracts can now work at their own pace without having to worry about ADAPT. Amazon did not respond directly when asked to confirm.
“Of course, some employees get scared by managers telling them they have to work faster, but if they are on permanent contract, Amazon cannot fire them,” said IP’s shop steward and Amazon warehouse employee in Sady, Agniezska Mroz.
Mroz, however, estimated that the ADAPT pause applies only to the 25% of Amazon employees with permanent contracts. The others can still be fired for low productivity.
Without ADAPT, Mroz said Polish Amazon had turned to the so-called Time-off-Task programme that adds up all the seconds when a worker is not scanning items. Too much “time off” can lead to punishment. Mayer of the Czech Dobrovíz warehouse also said that Amazon used Time-off-Task as a metric running parallel to ADAPT.
“The problem is that Amazon simply assumes you are slacking off when you could be talking to your manager or vomitting in the toilet,” said Mroz, speaking in the garden of Poznan’s popular anarchist Rozbrat Squat, its walls strewn with political graffiti and posters.
Amazon denied this, telling BIRN: “Performance is only measured when an employee is at their station and logged in to do their job. If an associate logs out, which they can do at any time, the performance management tool is paused. Meaning, if an associate needs to leave their station to wash their hands, refill their water, talk to a manager or use the restroom, to name just a few, they can do so once they are logged out without impacting their performance.”
In June, the company said that it would start averaging its measure of Time-off-Task over a longer period.
BOX 3: Where has all the data gone?
A major question concerning algorithmic management in Amazon is where worker data is stored and processed and whether it is in line with EU data protection regulations, notably the GDPR and a July 2020 ruling by the European Court of Justice that invalidated the so-called EU-US privacy shield allowing personal data transfer from the EU to the US (the Schrems II decision).
The Luxembourgish National Commission for Data Protection, CPND, told BIRN that Amazon shares personal data with its US headquarters on the basis of so-called standard contractual clauses. Standard contractual clauses were not invalidated by the Schrems II case. However, they must ensure adequate protection of personal data in the US. This can be individually reviewed by national supervisory authorities and suspended if such protection cannot be ensured. The CPND, which oversees Amazon’s European headquarters, has not reviewed Amazon’s standard security clauses regarding workers’ data since the Schrems II case. Following recent Politico revelations regarding protection of Amazon customers’ data, the CPND has been criticised for exercising excessive leniency vis-à-vis Amazon and other tech giants it oversees. Nevertheless, in July 2021, CPND pleased EU data watchdogs when it hit Amazon with an initial fine of 746 million euros for misusing customer data in targeted advertising.
In the privacy notice presented to candidates for employment, Amazon says it shares personal data with a list of Amazon affiliates, including both the European and US branches of Amazon Web Services and ‘shared services entities’, including a branch called Amazon Data Services Czech Republic.
Amazon Data Services Czech Republic was established in 2014. Since then, according to its financial statements, it has not had any salaried employees. Its executive director, Mr. Petr Lulák, runs a company specialising in creating ready-made legal persons, including in tax havens. Currently, he has roles on the executive boards of numerous firms. He could not be reached for comment regarding his involvement in Amazon Data Services Czech Republic. The Amazon Web Service spokesman for Central and Eastern Europe, Robert Belle, did not comment directly on the role of Amazon Data Services Czech Republic in the handling of personal data of workers.
Part of the culture
Any effect trade unions may have on how Amazon uses algorithmic management may have a ripple effect on other companies in the region.
Indeed, Amazon’s algorithmic management, while sophisticated, is part of a work culture that has existed for some time in Central and Eastern Europe and that is likely to be appropriated by emerging e-commerce companies operating in the region.
The automotive industry, the backbone of the Czech economy and a vital employer across the region, has long worked according to practices of “lean management” or the “just-in-time principle”, with the aim of cutting out inefficiencies and delays. Technical management, whereby the movement and pace are dictated by assembly lines, is indispensable in this industry.
“With conveyor belts, management also knows where you are, all the time,” said Sebastiano Lora, a former Amazon employee currently working in logistics.
Journalist Sasa Uhlova, who spent six months working in low-paid agency jobs equipped with a camera for the 2018 project The Limits of Work, said that like in Amazon, time pressure was constant and often counter-productive in these kind of jobs.
“There is a culture of self-exploitation and huge peer pressure,” Uhlova told BIRN. “It is everyone’s highest goal to be fast, or else they will call you a slacker. Nobody has time to question the system.”
Workers have offered similar accounts from the likes of Czech grocery delivery startup Rohlik.cz.
Some have described leader boards that show how many items individual workers have picked up, with the winners offered financial rewards and the losers threatened with termination.
“It was tough to remain in green numbers [above average], even as a young, fit man,” said one former employee at Rohlik.cz who spoke on condition of anonymity. “During one six-hour shift, I would walk up to 25,000 steps.”
Rohlik.cz told BIRN it was improbable that a worker would walk so far given the size of their warehouses. It said that workers are subcontracted via external agencies which set productivity rates and evaluate them.
With the growth in online shopping and next-day delivery becoming almost standard, ‘smart management’ of warehouses appears unavoidable. But unions say smart devices can be employed in ways more favourable to employees.
“We are not necessarily against algorithmic management, if done with transparency and in the context of collective bargaining,” said a spokesperson for UNI Global Union, a transnational union representing employees in commerce and logistics.
“I believe this is a huge issue in Amazon, where workers don’t have a say into the algorithms, and employees report that management by algorithm is having life-changing consequences including stress and brutal production rates.”
Even the most motivated employees may find it hard to keep pace.
“I know everything there is to know about this position,” said David Kopp, like Makova a resident of Most who loads trucks for Amazon. But with the increase in demand during the COVID-19 pandemic, Kopp said he had been unable to take in enough fluids or make toilet breaks, leading, he said, to gout, a form of inflammatory arthritis.
“I hoped I could stay in Amazon until my retirement,” Kopp said. “But with these health impacts, I doubt this will be possible.”
By Klára Votavová, originally published by BIRN/Reporting Democracy. Republished with permission. Main illustration: BIRN/Igor Vujcic.