When work control runs through software

Systems range from tracking location and pace in logistics to indicators of connection, keystrokes and screen time in office work, by way of analysis of exchanges in contact centres.

What they share is continuity: measurement is no longer an occasional check but a permanent recording, which changes the nature of what is observed and not only its volume.

A second change concerns the use of the data. In some systems, the indicator no longer merely records: it directly triggers an instruction, an alert or an allocation of tasks. One then speaks of algorithmic management, and no longer merely of surveillance.

The European data protection regulation requires a legal basis, a determined purpose, minimisation of the data collected and prior information of the people concerned.

The artificial intelligence regulation adopted in 2024 classes systems used for worker management among high-risk uses, imposing documentation and human oversight obligations whose application is phased.

The platform work directive, adopted the same year, contains provisions specific to algorithmic management, including a right to obtain an explanation of a significant automated decision. Its transposition leaves member states some latitude over the details.

The labour code requires control means to be proportionate to the aim pursued, brought to employees' knowledge beforehand and submitted for consultation with staff representatives.

The data protection authority has issued several decisions recalling that permanent and disproportionate surveillance is not lawful, including where a legitimate purpose is invoked. The practical difficulty lies in assessing proportionality, which is done case by case.

Available work finds associations between perceived surveillance and indicators of strain at work, with substantial moderation by the transparency of the system and by the possibility of contesting it.

Effects on output are more uncertain. Several studies find a rise in measured indicators accompanied by a fall in unmeasured behaviours, notably mutual help, a mechanism long known as a displacement effect. These results share a common limit: the organisations willing to open their data to researchers are not representative, and the systems studied evolve faster than the publications.

An activity indicator, keystroke time or connection, measures a trace and not work. It captures poorly the tasks of thinking, coordinating and preparing, which leave no comparable trace.

That limit is the same long encountered in management: what is easy to count takes the place of what counts, and automating measurement amplifies that displacement. Defenders of these tools reply that they make objective assessments once left to a superior's judgement, an argument that presupposes the chosen indicator is relevant.