Nearly one in five EU employees report that a program monitors their work performance. Eurofound published in 2026 an analysis of algorithmic management in European workplaces based on EWCS 2024 data. It distinguishes two uses that do not have the same effects: task allocation by algorithm, and performance monitoring by algorithm. Algorithmic performance monitoring is associated with greater exhaustion, particularly when work is already under strain; the first, depending on deployment conditions, can remain neutral or even relieve certain organizational frictions.
Key Points
- Algorithmic management affects nearly one in five European employees: 19% report performance monitoring, 18% algorithmic task allocation (Eurofound, 2024).
- The two uses do not have the same effects: algorithmic performance monitoring is associated with more unfavorable effects in jobs already under strain; the report recommends information, consultation, and safeguards.
- Exposure spans very different sectors, from transport to healthcare, passing through logistics and retail, with highly unequal exposure levels across countries and job categories.
- The 2026 report stresses that the prevalence and forms of algorithmic management vary between member states and types of employment.
- Deploying management AI without creating constant surveillance is possible, but this requires rules on transparency, negotiation, and recourse that few companies have yet implemented.
Nineteen Percent—And This Figure Masks Two Distinct Realities
Nineteen percent of EU employees report that a program monitors their work performance. Eighteen percent report that a program allocates their tasks. These two proportions look similar, but they describe very different experiences.
Allocating tasks by algorithm is what a delivery platform does when it sends an order to the nearest courier, or what hospital scheduling software does when it builds nursing rotations. The algorithm organizes. It can be constraining, but its action happens before the employee begins to work.
Monitoring performance by algorithm is something else. The program observes in real time: number of calls handled, time between two actions, movements in a warehouse, package scanning pace, breaks. It measures continuously. According to Eurofound, algorithmic performance monitoring is the function most strongly associated with adverse effects on health and well-being. In jobs where demands exceed resources, algorithmic performance monitoring adds to existing strain.
Exhaustion is reported more frequently by employees subject to algorithmic performance monitoring and facing high workplace strain.
This distinction matters because it prevents a binary reading. Algorithmic management is not uniformly bad. Task allocation can reduce informal injustices: the manager who favors their preferred employees, arbitrary scheduling of difficult shifts. The algorithm can introduce a form of neutrality. But this neutrality has its own biases, and it disappears as soon as one shifts from allocation to monitoring.
The Most Exposed Sectors Are Not the Ones People Imagine
Common depictions of algorithmic management concentrate it in digital platforms: bicycle couriers, rideshare drivers, remote workers monitored by productivity software. This image is accurate but incomplete.
Eurofound shows that exposure spans very varied sectors. Logistics and warehousing come first: Amazon warehouses and those of its European competitors have widely documented the use of systems measuring individual pace in real time. But retail commerce, call centers, and even certain segments of healthcare appear in the data. In nursing homes in some Nordic countries, algorithmic scheduling applications organize care worker routes and measure time spent with each resident.
Geography matters. Exposure rates vary significantly across countries. Economies with high intensity in digital services and logistics—the Netherlands, Ireland, the Baltic states—show higher proportions. Economies where collective bargaining still covers a large majority of employees—Scandinavia, Germany in certain sectors—have more closely supervised deployments, without stopping them entirely.
What stands out from Eurofound’s mapping is that algorithmic management is no longer a niche curiosity reserved for startups and platforms. It has become established in ordinary companies, with ordinary human resources, ordinary unions, and employees who don’t necessarily have the tools to understand what is measuring them.
The Algorithm Measures: Who Controls It
The information asymmetry between employer and employee is not limited to opacity about parameters. It also concerns the timing of measurement: the system records continuously, while the employee perceives its effects only later, at a meeting or when a decision that seems to flow from subjective judgment is actually the result of data aggregated over weeks. This desynchronization prevents any real-time correction and makes contestation difficult, because the employee lacks the traces that would allow them to reconstruct the sequence leading to the evaluation. Effective control therefore requires not only a right of access to data, but access within a timeframe and in a form that make understanding truly practicable—something current systems do not guarantee.
The control question is what the Eurofound data reveal. The report establishes an association between algorithmic performance monitoring, workplace strain, and reported exhaustion.
Negative effects vary by algorithmic function and are particularly marked for performance monitoring among employees already under workplace strain. The report recommends transparency and safeguards, while the exhaustion observed is associated with performance monitoring and workplace strain.
Conversely, Eurofound identifies configurations where algorithmic management remains more acceptable. These are cases where worker representatives were involved in defining indicators and where there is a right to request human review of an algorithmic decision. For high-risk AI systems at work, the European AI regulation provides notably for prior information to worker representatives and obligations for risk management, documentation, human oversight, and monitoring for relevant operators. AI systems that monitor or evaluate work performance and behavior fall in principle into this category, but management tools that are not AI systems or that do not meet these conditions may not be subject to the most stringent regime.
AI jobs remain trapped in major cities: this geographic concentration of technical expertise reinforces asymmetry. Employees most exposed to algorithmic management are rarely those with resources to contest how it functions.
Achievements of Collective Bargaining
Several European experiences show that constraining algorithmic management through negotiation is possible and produces measurable results.
In Germany, works councils, whose co-determination rights date to the 1970s, have obtained in several major companies a right to prior information before any deployment of a performance monitoring system. Volkswagen, Deutsche Post, and several supermarket chains have signed agreements that set authorized indicators, their collection frequency, and conditions for managerial access to individual data. These agreements frame tools without prohibiting them.
In the Netherlands, the Works Councils Act provides for consultation rights and, in certain cases, works council consent on decisions affecting organization or conditions of employment, and courts have progressively included algorithmic deployments in this category. Several legal proceedings have led companies to modify or suspend surveillance systems deemed disproportionate.
These advances remain fragile and unevenly distributed. They concern mainly large companies with active worker representatives. SMEs, which employ the majority of European employees, are largely outside this scope. Subcontracting makes the picture worse: employees of subcontractors working under algorithmic systems deployed by their clients have even less leverage than direct employees.
The Rules Still Missing for 2026-2035
The fossilization of current practices into implicit norms constitutes the central risk for the coming period. When a management tool is deployed without formal framework for several years, it tends to be perceived as a neutral element of the organization, which makes its later questioning politically and legally more costly. Union actors who were not involved from the start in deployment find themselves having to contest a fait accompli rather than negotiate the conditions of introduction. This dynamic argues for early intervention by social partners, before practices stabilize, rather than corrective regulation applied to usages already rooted in the managerial culture of the companies in question.
Deploying algorithmic management in ordinary European companies poses a question without an established answer: the conditions for its compatibility with work quality over time remain to be defined.
The 2026 Commission report stresses that algorithmic management and digital surveillance can be associated with increased stress, workload and pace, as well as reduced autonomy. Some member states have legislated specifically; others rely on general directives that did not anticipate these uses. The European AI regulation provides a foundation; management tools that are not AI systems or that do not meet the planned conditions can escape the high-risk regime, while AI systems for performance monitoring or evaluation fall into it in principle.
Three action areas emerge for the coming period. The first is algorithmic transparency at work: that employees know what measures them, with what parameters, and for what purpose. GDPR constrains certain fully automated decisions and guarantees notably, in applicable cases, the right to obtain human intervention, express one’s point of view, and contest the decision; its exercise at work remains poorly documented and little practiced.
The second action area is sectoral negotiation. Company agreements like those negotiated in Germany are useful but limited: they require strong union representation. Sectoral agreements, like those that exist in European metalworking or chemicals, could extend these protections to sectors like logistics or retail where representation is weaker but exposure is stronger.
The third action area is technical: designing algorithmic management systems whose indicators are verifiable, contestable, and revisable. Several researchers in computer science and ergonomics are working on architectures that make algorithmic decisions auditable by third parties—unions, labor inspectors, and judges. Transparency must be embedded in the software structure itself, not reduced to a declaration of good intent.
Whether this third action area advances fast enough depends partly on regulatory pressure, but also on companies that decide to get ahead of it. A few actors, notably in distribution and healthcare, have begun communicating about their commitments regarding algorithmic HR governance, partly to attract and retain employees in demographically strained sectors. This signal is still a minority view, but it exists.
European social partners will need to define, in the coming years, forms of democratic control over management tools that will allow productivity to be maintained without transforming work into continuous surveillance, before current practices fossilize into implicit norms.
