The Warehouse and the Shop Floor: Pace, Rate and Monitoring

In a modern fulfilment centre the instruction about what to do next, and how fast, comes from a hand-held scanner. Each scan records an action, the actions add up to a rate, the rate is compared with a target, and the comparison can produce a warning before any supervisor has seen the worker. This is one of the best-documented forms of the AI boss, and it shows most clearly what happens when the system that sets the pace of work is not the person who sees what the pace does to the body doing it. This chapter describes how rate and time-off-task systems direct and evaluate paced work, sets out what reporters, a union coalition, a United States Senate committee, the French data protection authority and its highest administrative court, and the California legislature found about them, and connects the findings to the employer's duty under European occupational safety and health law to adapt work to the individual. It then follows the pattern into home-care visits and claims queues and ends with how pace can be designed with the people who carry it. The chapter concerns three management functions: M1 Direction, because the rate system assigns, sequences and paces the work; M2 Evaluation, because the same data rates each person against the target; and M5 Care, because a system that sets the pace takes over the variable on which safety and health most depend.

The chapter builds on four conclusions of the earlier volumes. Human oversight must be designed, not declared, and approvals given in seconds are a signal of rubber-stamping (F21; AI Governance, Chapters 10 and 21) (41, Nordqvist, 2026). Approval quality is measured by time to approval, rejection rate and reversals, and where works councils have rights those metrics are agreed first and reported in aggregate (F52; AI Red Team, Chapters 6 and 15) (43, Nordqvist, 2026). Skills erode when AI does most of the work, yet the degraded mode after a stop needs them (F53). Reliance on automation reduces checking, and the response includes an oversight-quality metric reported at team level (F66; AI Threat, Chapters 6 and 14) (44, Nordqvist, 2026). It takes up Q11, what replaces the manager's judgement in the duty of care, and contributes evidence to Q4, whether algorithmic management works when measured in health as well as output.

Figure 6.1: The pace loop of target, measurement, warning and consequence, with the points where the safety representative, the risk assessment, the human reviewer and the named decider intervene.

Rate and time off task

A rate system has four parts. The target states how many units of work a person should complete in a period, such as items picked per hour or claims closed per day. The measurement comes as a by-product of directing the work, because the scanner that tells the picker which bin to go to next also records when the item was scanned. The comparison turns measurements into a figure per person, usually a rate against the target and a measure of time off task, the minutes in which the device recorded no work. The follow-up attaches something to the figure, from a message on the screen to a warning or, at the end, termination.

Katherine Kellogg, Melissa Valentine and Angèle Christin describe six mechanisms of algorithmic control: employers direct workers by restricting and recommending, evaluate them by recording and rating, and discipline them by replacing and rewarding (137, Kellogg et al., 2020). The rate system performs all six, from issuing one task at a time to linking the rating to rewards and replacement. Alessandro Delfanti's study of Amazon's warehouses describes the same design from the floor: sensors and scanners that track workers' movement around the building, and technologies whose aim, in his account, is to standardise, measure and discipline human work rather than to replace it (140, Delfanti, 2021).

Three properties distinguish a rate system from the stopwatch of scientific management. First, the target is often set from the distribution of measured performance rather than from a study of the task. Amazon told The Verge in 2019 that it changed rates only when more than 75 per cent of workers at a facility were meeting the goal, and that the bottom 5 per cent were placed on a training plan (187, Lecher, 2019); in 2024 the company described its expectations as based on what the team at a site is actually accomplishing in a given week, adjusted for seasonality (212, Amazon, 2024). A target defined from what most people achieve moves when they achieve more, and it always places some people below it. Second, measurement is continuous and individual, so that every minute of every shift is evidence. Third, the follow-up can be automated, which is where M2 Evaluation becomes M3 Consequence without a person deciding.

The rate is therefore M1 Direction and M2 Evaluation in one artefact: the number that tells the worker how fast to go is the number by which the worker is judged. The governance questions of this chapter concern a parameter, the target, and a link, the path from a figure to a consequence, rather than a model.

What was reported

The public record is dominated by one company, because its scale, its litigation and the attention of legislators produced documents that other employers' systems have not. What follows separates what each source reports from what it has been taken to mean.

Automated warnings and terminations, 2019

In April 2019 The Verge published documents that Amazon had filed in a dispute before the United States National Labor Relations Board. According to the company, roughly 300 full-time associates at one fulfilment centre in Baltimore had been terminated for inefficiency between August 2017 and September 2018, at a site of about 2,500 full-time employees. A letter from the company's lawyer stated that the system tracked the rate of each associate's productivity and "automatically generates any warnings or terminations regarding quality or productivity without input from supervisors". Time off task was tracked too, and long breaks from scanning generated warnings. Amazon said that supervisors could override the process and that an appeal system was part of it (187, Lecher, 2019). Chapter 8 analyses the case as consequence by algorithm; for this chapter the point is that target, measurement and warning were joined without a step at which a person had to look at the work.

The injury analyses, 2021

In May 2021 the Strategic Organizing Center, a coalition of trade unions, published Primed for Pain, an analysis of the injury and hours data that Amazon submitted to the United States Occupational Safety and Health Administration through its Injury Tracking Application for 2017 to 2020 (197, Strategic Organizing Center, 2021). It reported that in 2020 there were 5.9 serious injuries per 100 Amazon warehouse workers, counting injuries that required time off work or light duty, against 3.3 for all other employers in the warehousing industry, nearly 80 per cent higher. The total recordable injury rate at Amazon warehouses rose from 7.5 per 100 workers in 2017 to 9.0 in 2019 and fell to 6.5 in 2020. The report linked the fall to the company's suspension in March 2020, according to its own legal filings, of warnings and discipline for underperformance on productivity metrics, and noted that revised rates returned in October 2020. Sortable fulfilment centres with mobile shelving robots had a serious injury rate of 7.9 per 100 workers in 2019 against 5.1 at comparable sites without them, and in an online survey of workers in February 2021 eight in ten respondents with serious pain or injury related it to production pressure or speed.

The method sets the limits. The injury figures are the company's own regulatory submissions, re-analysed by a party with a stated interest. The 2020 fall coincides with the suspension of rate enforcement, but also with every other change the pandemic brought, so it suggests rather than demonstrates a causal link. The survey respondents chose to respond.

The Senate committee, 2024

In December 2024 the majority staff of the United States Senate Committee on Health, Education, Labor and Pensions, under its chair Bernie Sanders, published the findings of an 18-month investigation under the title The "Injury-Productivity Trade-off": How Amazon's Obsession with Speed Creates Uniquely Dangerous Warehouses (202, U.S. Senate Committee on Health et al., 2024). According to the committee's announcement, its analysis of Amazon's data found that the company's warehouses recorded over 30 per cent more injuries than the warehousing industry average in 2023, that in each of the previous seven years Amazon workers were nearly twice as likely to be injured as workers in other warehouses, and that more than two-thirds of the company's warehouses had injury rates above the industry average. It found that the company's speed requirements made its safety procedures nearly impossible to follow, and that the company had studied the connection between speed and injuries, including in an internal project known as Soteria, and declined to implement injury-reducing changes for fear of lower productivity.

Amazon answered in a statement published on 16 December 2024 that the report was wrong on the facts and relied on selective and outdated information (212, Amazon, 2024). It stated that its recordable incident rate in the United States fell by 28 per cent from 2019 to 2023; that it benchmarks its large warehouses against industry employers with more than 1,000 employees; that its own economists had found Project Soteria flawed and that it was never used; and that a judge in Washington State had vacated citations issued by the state's Department of Labor and Industries, ruling that the department had not established that the pace of work at Amazon was hazardous.

Reading the record

The parties disagree about which comparison is fair and whether a disowned internal study shows what the company knew. What both sides accept is narrower and still significant: the company sets expectations of pace for individuals, measures each person continuously against them, and has examined the relation between those expectations and injury. Chapter 14 weighs the injury evidence with the occupational-health literature that explains the mechanism. For the management functions the conclusion does not depend on settling the dispute. When a system sets the pace, the question whether the pace is safe has become a question about the system's parameters, and someone has to own it.

The French decision

The clearest European decision on warehouse monitoring concerns the same company and turns on the same data. After inspections in November 2019 at two warehouses of Amazon France Logistique, the restricted committee of the CNIL, the French data protection authority, imposed a fine of EUR 32 million by deliberation SAN-2023-021 of 27 December 2023. On appeal, the Conseil d'État reformed the deliberation on 23 December 2025 and reduced the fine to EUR 15 million (204, Conseil d'État, 2025). Chapter 13 takes from the case the principle of retention by purpose; this chapter takes what it says about pace.

The CNIL had found that three indicators lacked a lawful basis under Article 6(1)(f) of the GDPR (5, European Union, 2016): one raised when less than 1.25 seconds separated two scans during stowing, one recording scanner inactivity of more than ten consecutive minutes, and one recording inactivity at the start and end of the day and around breaks. In its reasoning, as the Conseil d'État summarises it, continuous timing of the order of a second went beyond what employees could reasonably expect and infringed both their privacy and their right to working conditions that respect their health and safety. The Conseil d'État disagreed on the indicators. The fast-scan indicator applied to one of four direct tasks, was raised only when two scans came anomalously close together and served to detect handling errors, so it could not be regarded as imposing constraints of speed; the idle-time indicator was raised only after ten consecutive minutes and did not govern breaks. The court noted, however, that after such a flag employees might have to give reasons touching their private lives in answer to their supervisors' questions. It upheld the finding that keeping every indicator for every employee for 31 days, for several purposes, breached data minimisation.

Two lessons follow for anyone designing or negotiating a rate system. An indicator's effect on pace depends on what it is for and what follows from it: the same timestamps can feed a quality check that flags implausibly fast handling or a speed target that penalises slow handling, with opposite effects on the person scanning, and the court rewarded narrow definition by purpose, threshold and task, with breaks expressly outside it. And the conversation after the flag is part of the system. A ten-minute idle flag is harmless as a record and intrusive as an interrogation, and whether a person must explain a visit to the toilet to a supervisor is decided by the follow-up rule, not by the data.

Quotas made visible

California chose a different instrument. Assembly Bill 701 of 2021 added sections 2100 to 2112 to the California Labor Code, in force from 1 January 2022, for employers with 100 or more employees at a single warehouse distribution center or 1,000 or more across the state (82, California State Legislature, 2021). A quota is a work standard of speed, tasks or material within a defined period, failure to meet which may lead to adverse employment action (section 2100(h)).

The law does four things. It requires the employer to give each employee a written description of each quota, including the number of tasks or the quantity of material in the defined period and any adverse action that could follow from missing it (section 2101). It makes unlawful any quota that prevents compliance with meal or rest periods, use of bathroom facilities, including reasonable travel time to and from them, or occupational health and safety rules, and forbids adverse action for failing to meet such a quota or one that was not disclosed (section 2102). It treats any action an employee takes to comply with occupational health and safety rules as time on task and productive time for any quota or monitoring system (section 2103). And it gives a current or former employee who believes a quota caused a violation of those rights the right to the written description and to the most recent 90 days of their own work speed data, within 21 calendar days of the request (section 2104).

European law has no such statute; its nearest equivalents are the GDPR's right of access and the duty of an employer deploying a high-risk AI system to inform workers' representatives and affected workers before use (AI Act, Article 26(7)) (1, European Union, 2024). The Californian law is instructive because it treats the target as a document. A quota that must be written down, handed to the person it governs, tested against the right to rest and produced on request is a quota that can be versioned, reviewed and contested. That is the form this book asks for every parameter that sets the pace of work.

Pace and health

Why pace matters for health is the subject of Chapter 14, and only the core is needed here. Robert Karasek's demand-control model located mental strain where high job demands meet low decision latitude, the worker's authority over how and when to meet the demands (107, Karasek, 1979). A rate system acts on both variables at once: it raises demand when the target rises with the distribution of performance, and it lowers latitude because the order and speed of tasks are set by the device. The Swedish Work Environment Act names the result in one sentence: strongly controlled or restricted work shall be avoided or limited (Chapter 2, section 1) (75, Sveriges riksdag, 1977).

Monitoring affects performance as well as health, and not in one direction. Lamar Pierce, Daniel Snow and Andrew McAfee studied 392 restaurant locations that adopted a theft-monitoring system at staggered dates and found reduced theft and improved productivity that appeared to come from changed behaviour rather than dismissals (129, Pierce et al., 2015). Ethan Bernstein, in a field experiment at the second-largest mobile-phone factory in the world, in China, found that visibility led workers to conceal their activities and that a modest increase in group-level privacy sustainably improved line performance (122, Bernstein, 2012).

Paced and monitored work is a European condition, not only an American one. Drawing on its 2019 enterprise survey, the European Agency for Safety and Health at Work found that 12 per cent of European workplaces used systems that determine the content or pace of work and 8 per cent systems that monitor performance, with manual sectors such as manufacturing reporting more pace-setting systems, and experts naming healthcare, logistics and manufacturing as the likeliest adopters of AI-based worker management (86, European Agency for Safety and Health at Work, 2022). In its 2024 survey, 7 per cent of establishments in the EU-27 reported systems monitoring performance or behaviour, 15 per cent in transportation and storage, and only 35 per cent of establishments using digital technologies had consulted employees about their health and safety effects (88, European Agency for Safety and Health at Work, 2025).

The employer's duty

European occupational safety and health law was written for assembly lines, and it fits rate systems closely. Chapter 14 sets out the duties in full; three provisions of Council Directive 89/391/EEC, the Framework Directive, carry the argument here (49, Council of the European Communities, 1989). Article 6(2)(d) requires the employer to adapt the work to the individual, especially in the choice of working and production methods, "with a view, in particular, to alleviating monotonous work and work at a predetermined work-rate and to reducing their effect on health". Article 6(3)(b) requires the employer, when entrusting tasks to a worker, to take the worker's capabilities as regards health and safety into consideration. Article 6(3)(c) requires consultation on the planning and introduction of new technologies as regards their consequences for safety and health. The Charter adds every worker's right to working conditions which respect his or her health, safety and dignity (Article 31(1)) (48, European Union, 2012).

A target generated by software is therefore a predetermined work-rate, to be assessed and alleviated whoever designed it; a rate that applies one number to everyone needs a documented way to adapt to the person; and a change to the target or to what follows from missing it requires assessment and consultation before it takes effect.

This is the sense in which M1 Direction becomes M5 Care when the system sets the pace. A supervisor who set the pace by hand also saw the person who could not keep it. When pace is set by a parameter, nobody sees the person unless the system is designed to show them, and the duty of care has to be exercised over the parameter: its value, its adaptation to individuals, its recovery time and the path from a missed target to a consequence. For paced work, the answer to Q11 begins with treating the target as a safety-relevant setting owned by a named person and assessed with the safety representative.

Co-determination law gives workers' representatives a hold on the same settings. In Germany, BetrVG §87(1) no. 6 gives the works council co-determination over technical equipment intended to monitor behaviour or performance, and the Federal Labour Court reads "intended" objectively: equipment is covered when it is objectively suited to collect and record such information, whatever the employer's intention (9, Germany, 2001) (71, Bundesarbeitsgericht, 2016). Every scanner that records task times falls within it. Section 87(1) no. 7 adds co-determination on health protection; section 90 requires the employer to inform and consult the works council in good time on planned work procedures, now expressly including the use of artificial intelligence; and where changes manifestly contradict established ergonomic knowledge and place employees under particular strain, section 91 lets the works council demand measures, with the conciliation committee deciding if the parties disagree. A rate system engages all four provisions, and the health provisions are the ones most often overlooked. The Swedish model works through the safety representative, who may ask the employer for measures or an investigation and take the matter to the Work Environment Authority if the employer does not act (Work Environment Act, Chapter 6, section 6a) (75, Sveriges riksdag, 1977), and through the Authority's rule that the tasks given to workers must not give rise to unhealthy workload, so that resources are adapted to demands (AFS 2023:2, Chapter 2, section 6) (84, Arbetsmiljöverket, 2023). A rate is a demand; recovery time inside the rate is a resource.

The AI Act adds its own layer. It classifies as high-risk AI systems intended to allocate tasks based on individual behaviour or personal traits or characteristics, or to monitor and evaluate the performance and behaviour of persons in work-related relationships (Annex III, point 4(b)) (1, European Union, 2024). A fixed rule that counts scans may not use AI at all and still sets a predetermined work-rate. Under the Digital Omnibus the high-risk obligations, including the deployer duties of Article 26, apply to Annex III systems from 2 December 2027 (2, European Union, 2026), which is expected to bring rate systems in European logistics and manufacturing under deployer obligations for the first time. For platform work only, the Platform Work Directive already requires platforms to evaluate the risks of automated monitoring and decision-making systems to safety and health, "in particular as regards possible risks of work-related accidents, psychosocial and ergonomic risks", and not to use them "in a manner that puts undue pressure on platform workers" (Article 12) (47, European Union, 2024). This book uses that article as the benchmark for any paced work under algorithmic direction.

The pattern elsewhere

The same four parts, target, measurement, comparison and follow-up, appear wherever work arrives in units that software can count.

In home care, the unit is the visit. Sian Moore and L. J. B. Hayes studied electronic monitoring in English homecare, where carers log in and out of clients' homes and local authorities commission care by the minute. They argue that minute-by-minute commissioning anchors the cost of care in the time spent in clients' houses and squeezes out paid travel, time between visits, training and supervision, so that paid working time is minimised, unpaid time grows and care work intensifies (213, Moore & Hayes, 2017). The planned visit time works as a rate, with one difference that matters for M5 Care: the work inside the visit is care for another person, and a visit that runs long because a resident fell, was confused or needed to talk is not a deviation to be corrected but the work done well.

In claims and service centres, the unit is the case or the call. Phil Taylor and Peter Bain, studying the rapid growth of call centres in Scotland in the late 1990s, described the work under continuous electronic measurement and targets as "an assembly line in the head" (210, Taylor & Bain, 1999). A claims queue paced by an allocation system reproduces the pattern: the system decides which case comes next, the handling time is recorded, throughput per person can be computed, and the follow-up decides whether the figure is a planning aid or a judgement.

On the manufacturing shop floor, the line set the pace long before software did, and the evidence from Bernstein's factory shows that adding individual visibility to a paced line can reduce performance. What software adds is the individual figure and the automatic follow-up. In each setting, the extension of rate logic is observable where systems already count units, as in visit registration, contact-centre workforce management and claims allocation, and expected where agents begin to assign and time office tasks, as Chapter 10 describes.

Designing pace with people

The record of this chapter supports seven design rules. They do not prohibit rate systems; they make the parameters that set pace into governed artefacts and keep consequence with a person. The target is set with the people who carry it, through the safety representative or works council, before it takes effect. Recovery is part of the rate: breaks, travel between tasks, safety actions and the time a care task genuinely needs count as work, as California's section 2103 does for safety actions. The rate adapts to the individual, as Article 6(3)(b) of the Framework Directive requires. A warning is reviewed by a person with the authority and the time to look at the work, and, following F21 and F52, review time and the rate at which reviewers depart from the flag are measured and reported at team level. No consequence follows from a figure without a named human decider, because Level 4 Autonomous is never used for M3 Consequence and an automated warning that counts towards discipline is a step of that consequence. Strain is reported at team level alongside output. And there is a stop path outside the system: the line manager and the safety representative can suspend the rate for a team, which the team can then plan by hand only if it has kept the skill to do so (F53).

Practical guide: Assessing a pace or rate system as a work-environment risk with safety representatives

Purpose. To assess any system that sets, measures or enforces the pace of work as a work-environment risk before it is introduced or changed, so that the duties of the Framework Directive and national law are met for the parameters that actually set the pace.

Roles. The line manager who owns the work (accountable); the safety representative or works-council member responsible for health and safety; the occupational health service; the system owner; the data protection officer; and two or three workers who do the work.

1. Describe the loop. Write down the four parts of the pace loop for the system: what the target is and how it is computed, what is measured and how often, how the comparison is presented and to whom, and what follow-up is attached to a missed target.
2. Name the parameters. List every setting that changes pace: the target value or formula, planned task durations, travel or transition allowances, the time-off-task threshold, the review period, and any thresholds that trigger messages or warnings. Give each an owner.
3. Observe the work. The safety representative and the line manager spend at least one full shift with the work as it is done, and compare the planned durations with what the tasks require, including the tasks the system does not count.
4. Assess the hazards. For each parameter, assess physical risks (repetition, load, haste), psychosocial risks (time pressure, low latitude, isolation) and risks from the follow-up (fear of warnings, skipped breaks).
5. Set controls in the parameters. Prefer changing parameters to asking workers to cope: recovery allowances, minimum transition times, task minimums, individual adjustments, and a time-off-task rule that never governs breaks.
6. Decide the follow-up. State in writing what a missed target may and may not lead to. By default, a missed target opens a conversation about the work, never an automated warning, and individual figures are not used in pay or discipline without a separate decision under the delegation profile.
7. Agree the reporting. Choose team-level strain indicators, the reporting interval (monthly by default) and a minimum group size for any breakdown (ten by default), and agree them with the representatives before any figure is shown.
8. Record and link. Record the assessment as a versioned file linked to the system's delegation profile and decision record, and set triggers for reassessment: any change to a parameter in step 2, a contestation of a target, an injury linked to pace, or a model change by the supplier.

Definition of done. Every pace parameter has an owner, a documented value and a control; the follow-up rule is written; the safety representative has signed the assessment or recorded a dissent; strain indicators are reported at team level; reassessment triggers are set; and the result is linked to the delegation profile. Adapt the defaults to the collective agreement, the works agreement and national law, which may set stricter requirements or give the representatives a right of consent rather than consultation.

Worked example: A work-environment risk assessment of visit-time targets in the Strandvik home-care unit

Illustrative. Nordvik Region's home-care dashboards compare planned with actual time per visit and per route, using the registrations on staff phones and a forecasting model for visit durations, and show individual figures to the unit manager at Level 0 Advisory under the delegation profile DLG-NR-HCD-01. The routes themselves are drafted by ShiftWise. When an assistant nurse contested a route that left no time between two visits (CNT-NR-0001), Mikael Strand and the unit's safety representative recognised that the planned visit times played the part of the warehouse rate, and assessed them as a work-environment risk. This is revision 1 of the unit's home-care assessment WRA-NR-HC-01 and covers the visit-time targets and what the dashboard does with them; revision 2, which Chapter 14 shows, adds the routing constraints that followed from the same contest. The Region recorded its decision as AIDR-0632, within the commission AIDR-0600.

risk_assessment:
  id: WRA-NR-HC-01
  revision: 1
  date: 2027-02-24
  scope: Planned visit times and the time-per-visit dashboard, Strandvik home-care unit
  systems: [home-care performance dashboards, visit-duration forecasting model]
  delegation_profile: DLG-NR-HCD-01
  next_revision: {revision: 2, date: 2027-03-11, adds: minimum travel time and breaks after CNT-NR-0001, shown in Chapter 14}
  trigger: CNT-NR-0001
  decision_record: AIDR-0632
  management_functions: [M1 Direction, M2 Evaluation, M5 Care]
  legal_basis:
    - "Directive 89/391/EEC Art. 6(2)(d), 6(3)(b), 6(3)(c), 11"
    - "Work Environment Act ch. 2 s. 1; ch. 6 s. 6a"
    - "AFS 2023:2 ch. 2 s. 6"
  assessed_by: [unit manager Mikael Strand, safety representative, occupational health service]
  workers_consulted: 3
  observation: two full day shifts and one evening shift, January 2027
  pace_parameters:
    - id: P1
      parameter: Forecast visit duration per visit type
      owner: unit manager
      hazard: Forecast below the time care tasks need for some residents
      control: Planned time never below the care-plan minimum for the visit
    - id: P2
      parameter: Overrun shown per visit on the dashboard
      owner: unit manager
      hazard: Long visits read as underperformance; haste in care
      control: Overruns shown per route and week; individual view only on request, logged
    - id: P3
      parameter: Documentation time after each visit
      owner: unit manager
      hazard: Uncounted work pushed into breaks or after shift
      control: Five minutes per visit counted as planned work
  follow_up_rule:
    missed_planned_time: opens a conversation about the route and the resident
    automated_warnings: none
    use_in_pay_or_discipline: excluded
  strain_indicators:
    - overrun_share_by_route
    - missed_breaks
    - late_finishes
    - contested_routes
    - sick_leave_days
  reporting: {interval: monthly, level: unit, minimum_group_size: 10}
  stop_path: unit manager or safety representative may hide planned-versus-actual figures for the unit
  reassess_on: [change to forecasting model, change to P1-P3, new contestation, injury linked to pace]
  signed:
    unit_manager: true
    safety_representative: true
  next_review: 2027-08-31

The assessment is short because the decisions are in the parameters. P1 turns the forecasting model from a target into a proposal bounded by the care plan, which is the M5 Care judgement a nurse would have made. P2 removes the per-visit overrun from the view a manager sees first, because in care a long visit is often the work done well. P3 counts the documentation the dashboard did not see, and the follow-up rule repeats the exclusion Chapter 8 recorded for DLG-NR-HCD-01 and adds that no automated warning exists. The five minutes in P3 and the group size of ten are the Region's choices for this illustration; the collective agreement and local conditions may require others.

Across the degrees of delegation

Degree	M1 Direction in a rate system	M2 Evaluation	Where the danger lies
Level 0 Advisory	Proposes targets and plans that a manager sets	Figures shown to a manager, no rating	Figures read as judgements without a rule
Level 1 Assisted	Drafts targets; a manager adopts them	Flags deviations for a manager to review	Review becomes a click; measure review quality (F21)
Level 2 Supervised	Sets task order and pace; a manager approves plans	Rates persons; a person confirms each warning	Warnings confirmed in seconds; recovery missing from targets
Level 3 Bounded	Paces work within agreed limits on rest and transitions	Issues coaching messages within an agreed rule	Limits written once and not reassessed when the model changes
Level 4 Autonomous	Not recommended for pace without health limits	Automated warnings that count towards discipline	Level 4 in M3 Consequence, excluded by this book's rule

Rate systems are usually Mode 4 Built AI applications, often beside Mode 0 Ambient AI in licensed workforce-management software that computes similar figures unasked, and typically operate at Level 1 Assisted or Level 2 Supervised for M2 Evaluation; the danger arises when warnings or terminations follow automatically, which is Level 4 Autonomous in M3 Consequence. Rate monitoring, the reported automated terminations, the CNIL's decision and its reduction, and the California law are observable; the link between monitored pace and injury is reported in analyses of regulatory data whose methods and limits are stated above; the application of the AI Act's high-risk duties to rate systems from 2 December 2027 is expected; and the extension of rate logic to care visits and service queues is observable, while its extension to office work through agents that assign and time tasks is expected.

Nordvik Region and Hansa Mutual

In the Strandvik home-care unit, the visit times planned in ShiftWise routes and registered on staff phones play the part of the rate. CNT-NR-0001, the assistant nurse's contestation of a route that left no time between two visits, was the unit's first visible sign of pace pressure. Mikael Strand and the safety representative assessed the planned visit times as a work-environment risk in WRA-NR-HC-01 revision 1, and the dashboards stay at Level 0 Advisory under DLG-NR-HCD-01, with no automated warning and no use of individual figures in pay or discipline. Because every workplace has a safety representative with the right to request an investigation, the contestation entered the work-environment system rather than remaining a complaint about software.

At Hansa Mutual the paced work is in the claims queues, which resemble a warehouse more than an office most of all at the Polish service centre, where about 1,400 of the group's claims staff work. In the German claims teams the claims-triage agent allocates claims to handlers at Level 3 Bounded under DLG-HM-CTA-01, a profile recorded for Germany only, which sets the order and, through the volume it releases, the pace of each handler's day. At the Polish service centre allocation does not run under that profile: team leaders allocate from the queues, and any extension of the agent's allocation there must first pass the Polish entity's own consultation gate (Chapters 4 and 25). Throughput figures per handler go to team leaders only, under the rule of the group works agreement KBV-IT-2024, which the group applies as policy in Poland, where a German works agreement has no legal force of its own; the Polish entity sets the purposes, scope and manner of the monitoring in its work regulations, as Article 22³ of the Labour Code requires, and informs staff at least two weeks before a change (79, Rzeczpospolita Polska, 1974).

Neither organisation uses automated warnings, and both record that as a decision rather than an absence: Nordvik in the follow-up rule of WRA-NR-HC-01 and the conditions of DLG-NR-HCD-01, Hansa in the excluded uses of DLG-HM-CTA-01. They differ in where pace is felt: in Nordvik in the time a resident needs, in Hansa in a queue that never empties, where the risk is a steady intensity that no single figure reveals and team-level strain indicators matter more than individual throughput.

Summary

Rate and time-off-task systems are the most developed form of the AI boss in paced work. A target, a continuous measurement, a comparison and a follow-up combine M1 Direction and M2 Evaluation in one number, and when the follow-up is automated they reach M3 Consequence without a person deciding. The documents published in 2019 described exactly that.

The injury evidence is strongest for one company and disputed by it. The French case showed that scanner data can be lawful as a narrow quality check and unlawful as a store of everything, and the California law made the quota a written, disclosed and contestable document. European law already requires employers to alleviate work at a predetermined work-rate, adapt it to the individual and consult before introducing new technology, and co-determination law gives works councils and safety representatives a hold on the settings that set the pace.

When the system sets the pace, M1 Direction becomes M5 Care: the duty of care has to be exercised over parameters, not only over people seen at work. The practical guide and worked example show how a line manager and a safety representative assess those parameters, build recovery into the rate, write down what a missed target may lead to and report strain at team level, in home care and claims queues as much as in warehouses.

Chapter 7 turns from the pace of work to access to it: hiring by machine, where screening, scoring and automated interviews decide who is employed at all, and where M3 Consequence begins before the employment relationship does.

Sources

1. Nordqvist, G. (2026). AI Governance: Governing Artificial Intelligence in European Enterprises and Public Authorities. The Architecture as Code Series, Book 2. (41, Nordqvist, 2026)
2. Nordqvist, G. (2026). AI Red Team: AI as an Active Part of Cyber Defence in European Enterprises and Public Authorities. The Architecture as Code Series, Book 4. (43, Nordqvist, 2026)
3. Nordqvist, G. (2026). AI Threat: Are the Dystopias Coming True? Evidence, Risk and Response for European Enterprises and Public Authorities. The Architecture as Code Series, Book 5. (44, Nordqvist, 2026)
4. Kellogg, K. C., Valentine, M. A. & Christin, A. (2020). Algorithms at Work: The New Contested Terrain of Control. Academy of Management Annals, 14(1), 366–410. (137, Kellogg et al., 2020)
5. Delfanti, A. (2021). The Warehouse: Workers and Robots at Amazon. Pluto Press. (140, Delfanti, 2021)
6. Lecher, C. (2019). How Amazon automatically tracks and fires warehouse workers for 'productivity'. The Verge. (187, Lecher, 2019)
7. Amazon (2024). Senator Bernie Sanders continues to mislead the American public on workplace safety at Amazon. About Amazon, 16 December 2024. (212, Amazon, 2024)
8. Strategic Organizing Center (2021). Primed for Pain: Amazon's Epidemic of Workplace Injuries. SOC. (197, Strategic Organizing Center, 2021)
9. U.S. Senate Committee on Health, Education, Labor and Pensions, Majority Staff (2024). The "Injury-Productivity Trade-off": How Amazon's Obsession with Speed Creates Uniquely Dangerous Warehouses. United States Senate. (202, U.S. Senate Committee on Health et al., 2024)
10. Conseil d'État (2025). Décision n° 492830 du 23 décembre 2025, Société Amazon France Logistique. ArianeWeb. (204, Conseil d'État, 2025)
11. European Parliament and Council of the European Union (2016). Regulation (EU) 2016/679 (General Data Protection Regulation). Official Journal of the European Union. (5, European Union, 2016)
12. California State Legislature (2021). Assembly Bill No. 701, Chapter 197, Statutes of 2021: Warehouse distribution centers (California Labor Code §§ 2100–2112). California Legislative Information. (82, California State Legislature, 2021)
13. European Parliament and Council of the European Union (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union. (1, European Union, 2024)
14. Karasek, R. A. (1979). Job Demands, Job Decision Latitude, and Mental Strain: Implications for Job Redesign. Administrative Science Quarterly, 24(2), 285–308. (107, Karasek, 1979)
15. Sveriges riksdag (1977). Arbetsmiljölag (1977:1160) (Work Environment Act). (75, Sveriges riksdag, 1977)
16. Pierce, L., Snow, D. C. & McAfee, A. (2015). Cleaning House: The Impact of Information Technology Monitoring on Employee Theft and Productivity. Management Science, 61(10), 2299–2319. (129, Pierce et al., 2015)
17. Bernstein, E. S. (2012). The Transparency Paradox: A Role for Privacy in Organizational Learning and Operational Control. Administrative Science Quarterly, 57(2), 181–216. (122, Bernstein, 2012)
18. European Agency for Safety and Health at Work (2022). Artificial intelligence for worker management: an overview. EU-OSHA. (86, European Agency for Safety and Health at Work, 2022)
19. European Agency for Safety and Health at Work (2025). First findings of the Fourth European Survey of Enterprises on New and Emerging Risks (ESENER 2024). EU-OSHA. (88, European Agency for Safety and Health at Work, 2025)
20. Council of the European Communities (1989). Council Directive 89/391/EEC of 12 June 1989 on the introduction of measures to encourage improvements in the safety and health of workers at work. Official Journal of the European Communities. (49, Council of the European Communities, 1989)
21. European Union (2012). Charter of Fundamental Rights of the European Union. Official Journal of the European Union. (48, European Union, 2012)
22. Federal Republic of Germany (2001). Betriebsverfassungsgesetz (BetrVG) (Works Constitution Act). Gesetze im Internet. (9, Germany, 2001)
23. Bundesarbeitsgericht (2016). Beschluss vom 13. Dezember 2016, 1 ABR 7/15. Bundesarbeitsgericht. (71, Bundesarbeitsgericht, 2016)
24. Arbetsmiljöverket (2023). AFS 2023:2 Planering och organisering av arbetsmiljöarbete. Swedish Work Environment Authority. (84, Arbetsmiljöverket, 2023)
25. European Parliament and Council of the European Union (2026). Regulation (EU) 2026/1744 (Digital Omnibus on AI). Official Journal of the European Union. (2, European Union, 2026)
26. European Parliament and Council of the European Union (2024). Directive (EU) 2024/2831 on improving working conditions in platform work. Official Journal of the European Union. (47, European Union, 2024)
27. Moore, S. & Hayes, L. J. B. (2017). Taking worker productivity to a new level? Electronic Monitoring in homecare: the (re)production of unpaid labour. New Technology, Work and Employment, 32(2), 101–114. (213, Moore & Hayes, 2017)
28. Taylor, P. & Bain, P. (1999). 'An assembly line in the head': work and employee relations in the call centre. Industrial Relations Journal, 30(2), 101–117. (210, Taylor & Bain, 1999)
29. Rzeczpospolita Polska (1974). Ustawa z dnia 26 czerwca 1974 r. Kodeks pracy (Labour Code). (79, Rzeczpospolita Polska, 1974)