TL;DR
- Eurofound measured three algorithmic management functions with the 2024 European Working Conditions Survey, 36,644 face-to-face interviews across 35 countries and the analysis restricted to the EU-27: 19% of employees are monitored by software, 18% have their tasks allocated by it, 10% have their working hours set by it.
- Performance monitoring is the function most consistently tied to worse health and well-being, and it adds to strain in jobs that are already strained.
- The largest meta-analysis on the subject, 94 independent samples and 23,461 people, reports a corrected correlation of .00 between electronic performance monitoring and task performance, and .16 with stress.
- The most widespread function in European workplaces is therefore the one with no measured productivity return and a measured human cost.
- What monitoring reliably changes is tempo: 44% of monitored employees say performance targets dictate the pace of their work, against 28% of everyone else.
Assumptions
- Eurofound reports associations from employee self-reports, not causal effects, and says so explicitly. Where a system is invisible to the worker, real prevalence is higher than the measured 19%.
- The meta-analysis pools monitoring tools studied over two decades, most of them older than the current generation of software. Anyone claiming the new tools behave differently has not published the numbers.
- The 1916 statute quoted below is a primary source I read in the Statutes at Large. The parallel I draw from it is mine, not the authors'.
- Vendors hold the only data on their own products' effect on output, and none of it is public.
Goal
To show why the productivity story told about workplace monitoring does not survive contact with the evidence, and what a buyer or a works council should ask instead.
The official version
Eurofound announced Algorithmic management in European workplaces on 29 October 2026, and the coverage settled on one word: surveillance. One in five European employees watched by software, Big Brother reborn, pick your headline.
Inside companies the story runs the other way and sounds more respectable, because monitoring is sold as an instrument of productivity. You cannot improve what you cannot measure, so measure everything and output will follow.
Both versions share an assumption nobody audits, which is that the thing works.
What the evidence says
Daniel Ravid and four colleagues pooled 94 independent samples and 23,461 workers in Personnel Psychology in 2023. Corrected correlation between the presence of electronic performance monitoring and task performance: .00, confidence interval from minus .06 to .06. Their own sentence is blunter than anything I would write: "Results provide no evidence that EPM improves worker performance."
The same analysis finds .16 with stress and strain, interval .13 to .20, and .22 between more invasive monitoring and counterproductive behaviour at work, that second interval starting at zero and so the weaker of the two claims. The intervention has no measured effect on the outcome it is bought for, and two measured effects nobody is buying.
Now set the Eurofound numbers beside that. Of its three functions, monitoring is the most widespread at 19%, and it is the one most consistently linked to exhaustion, work-family conflict and lower engagement. Automated task allocation sits at 18% and is the one function associated with higher engagement, and the authors spend a paragraph warning that the association probably reflects the kind of workplace that adopts it rather than any virtue of the tool. Honest of them, and ignored by everyone who quoted the report.
The most common function in European workplaces is the one with no productivity return in the literature and the clearest health cost.
What it does instead
Monitoring decides nothing. It issues no instruction, denies no leave, fires nobody, which is exactly why it travels so easily through a legal system built around automated decisions and personal data.
What it does is set tempo. Among monitored employees, 44% say performance targets dictate their pace, against 28% of the others. Nick Turner and Justin Weinhardt of the University of Calgary, writing on 29 July 2026, got the management problem into one line: "seeing more of work does not always mean understanding it better." They cite production lines in a phone factory screened off by curtains, which turned out to have fewer defects than the visible ones.
Cory Doctorow reached the human end of this on Democracy Now on 25 June 2026: "You're not just being used by the machine, you're being used up by it." He is describing extraction. The evidence points at something less cinematic and more embarrassing, because the wear is real and nothing arrives at the other end.
1916
Congress has been here before. The Navy appropriations act approved on 29 August 1916 carries a proviso denying pay to any officer or foreman "while making or causing to be made with a stop watch or other time-measuring device a time study of any job of any such employee". Workers at the Watertown Arsenal had walked off the job over Taylor's stopwatch in August 1911, five years earlier.
It also changed nothing about the pace of American industry, because Ford had already put the work on a moving chain in 1913. No stopwatch, no time study, no device pointed at any individual. The line set the speed, and the statute had nothing to say about a line.
The strongest objection to the parallel is that European law does name this. Annex III of the AI Act puts AI used to manage work relationships and to evaluate the performance and behaviour of workers in the high-risk class, with the documentation, data governance and human oversight that follow. Those obligations attach to the system and its decisions. None of them puts a ceiling on the pace the system sets, and a monitoring dashboard with no AI in it falls outside the Act to begin with.
Why this matters for your business
If you are buying one of these systems, or defending one you already bought, three of the four useful questions cost nothing:
- What is the measured effect on output, in your own data, against an unmonitored comparison group?
- Which of these numbers reach a manager, and which reach the person whose pace they set?
- Does any target in the system adjust itself automatically to the measurements it collects?
- If the answer to the first question is "we have not measured it", what is actually being managed here?
In a works council negotiation the data clauses are already table stakes. The clause worth spending political capital on is the one saying a target cannot be reset by the system that measures it.