SWITCHED ON
The daily technology series nobody asked for but everyone needed
The Boss That Never Blinks: Workplace Surveillance and the Productivity Metric Problem
Software that logs your every keystroke, screenshots your screen every few minutes, and scores your "productivity" against your colleagues has moved from fringe practice to mainstream deployment across millions of remote and hybrid workers. The evidence for whether it actually improves anything is considerably weaker than its adoption rate suggests.
The shift to remote and hybrid work removed the physical presence that managers historically relied on, consciously or not, as a proxy for productivity. Rather than developing better proxies for what actually matters — output, quality, outcomes — a significant portion of the market responded by trying to digitally recreate the physical presence itself, down to the individual keystroke. This is a category error with real consequences for the people being monitored.
Yesterday we covered AI-driven materials discovery — GNoME's staggering predicted materials catalogue, self-driving labs closing the validation gap, and the honest recognition that discovery compressing dramatically has not yet compressed the slower stages of manufacturing and deployment that follow it. Today we are addressing something considerably more immediate for a large fraction of this series' readers: workplace surveillance technology, colloquially known in some corners of the internet as "bossware." The monitoring software deployed to track remote and hybrid worker activity, what it actually measures, what the evidence shows about its effects, and the genuinely difficult underlying problem — measuring knowledge work productivity — that this technology has, in most documented cases, failed to solve while creating real costs of its own.
01 — What Is Actually Being Deployed
Employee monitoring software has expanded dramatically in both sophistication and adoption since the mass shift to remote work beginning in 2020, and the current generation of tools captures a considerably broader range of activity than the relatively basic time-tracking software that preceded it. Screenshot capture, taking periodic (sometimes as frequent as every few minutes) screenshots of an employee's screen throughout the working day, is among the most common features in widely deployed tools including Hubstaff, Time Doctor, and ActivTrak. Keystroke and mouse activity logging tracks the volume and pattern of computer interaction, often used to generate an "activity score" or "idle time" metric that managers can review. Application and website usage tracking monitors which programs and sites an employee has open and for how long, frequently categorised into "productive" and "unproductive" buckets by pre-set or administrator-defined rules.
More advanced deployments incorporate webcam monitoring (periodic or continuous capture confirming an employee's physical presence at their desk), and in some documented cases, AI-driven sentiment or engagement analysis applied to written workplace communications in tools like Slack and email, attempting to infer employee morale or disengagement from language patterns — a category of monitoring that raises the same fundamental accuracy and interpretive concerns we discussed regarding AI mental health and sentiment analysis tools in earlier seasons, now applied in a context with direct employment consequences attached to the inferences drawn.
A 2023 Gartner survey found that the proportion of large employers using some form of employee monitoring software had risen from roughly thirty percent pre-pandemic to over sixty percent by 2022, with continued growth since — a rate of adoption for a genuinely invasive workplace technology that occurred with remarkably little public debate relative to its scale, compared to the sustained scrutiny applied to comparable consumer surveillance technologies covered elsewhere in this series.
02 — The Measurement Validity Problem
The foundational problem with activity-based monitoring metrics — keystrokes, mouse movement, active application time — is that they measure activity, not output or value, and the relationship between the two for knowledge work is considerably weaker and more variable than for the kind of manual, repetitive labour where activity-based metrics originated (assembly line output counts, call centre calls handled per hour).
Genuine knowledge work frequently involves periods that look, by activity metrics, indistinguishable from idleness — reading and thinking without typing, discussing a problem verbally with a colleague, working through a difficult concept while staring at a wall — that are not merely compatible with high-quality output but are frequently prerequisite to it. Conversely, high measured activity — continuous typing, constant application switching, rapid-fire messaging — can reflect genuine productive engagement or can reflect a worker performing "productivity theatre," deliberately generating activity signals to satisfy a monitoring system's metrics regardless of whether that activity produces any actual value, a documented and increasingly discussed adaptation that workers under monitoring have developed, including software tools specifically designed to simulate mouse movement and keep monitoring software from flagging idle time during legitimate breaks or thinking periods.
Research on this dynamic, including studies published in the Journal of Business Ethics and MIT Sloan Management Review, has documented that monitored employees frequently adapt their behaviour to optimise for the metric being measured rather than for genuine output — a manifestation of Goodhart's Law (when a measure becomes a target, it ceases to be a good measure) that applies with particular force to knowledge work, where the gap between easily measurable proxies and genuinely valuable output is especially wide.
03 — What the Evidence Actually Shows About Effectiveness
The evidence base for whether monitoring software actually improves organisational productivity, as opposed to simply generating monitoring data that feels reassuring to managers, is notably thinner and more mixed than the scale of adoption would suggest, and the studies that do exist paint a genuinely complicated picture rather than a clean verdict in either direction.
Some studies, including research published by economists studying call centre and data entry environments where output is genuinely well-correlated with measurable activity, have found productivity improvements associated with monitoring, particularly for highly structured, repetitive tasks. Research examining monitoring's effects on more autonomous, judgment-intensive knowledge work has found more concerning results: a 2021 study published in the Journal of Applied Psychology found that monitored employees reported significantly higher stress and lower job satisfaction, with some evidence of increased "rule-breaking" behaviour as a documented psychological response to perceived surveillance and reduced autonomy — a finding consistent with a broader body of organisational psychology research on the counterproductive effects of low-trust monitoring environments on employee behaviour and retention.
A particularly instructive 2022 study by researchers at the University of Chicago and University of California, examining a large dataset of monitored remote workers, found that while monitoring did increase measured activity metrics, it showed no significant corresponding increase in actual completed work output, and was associated with measurable increases in employee turnover intention — a pattern consistent with the productivity theatre dynamic, in which employees learn to generate the activity signals the monitoring system rewards without a corresponding increase in genuine productive output, while experiencing the monitoring itself as a trust and autonomy cost that makes them more likely to leave.
04 — The Legal and Regulatory Landscape
The legal framework governing workplace monitoring varies considerably by jurisdiction and has been evolving, generally in the direction of greater disclosure requirements rather than outright restriction, reflecting a broader pattern in employment law of treating monitoring as a legitimate management prerogative subject to procedural safeguards rather than a practice requiring substantive limitation.
The EU's General Data Protection Regulation, which we covered in detail in Season One's surveillance economy episode, applies to workplace monitoring data as personal data, requiring a legal basis for processing, proportionality (monitoring should be no more invasive than necessary for its stated purpose), and in several EU member states, specific additional requirements including works council consultation before implementing new monitoring systems — a meaningfully more restrictive framework than exists in the US, where employee monitoring is governed primarily by a patchwork of state laws, with only a minority of states (including Connecticut, Delaware, and New York) requiring employers to provide advance notice before implementing electronic monitoring, and no comprehensive federal framework specifically addressing workplace surveillance technology.
This regulatory divergence has produced a genuinely different practical monitoring landscape across jurisdictions — European employers face meaningfully higher compliance burdens and more employee consultation requirements before deploying invasive monitoring tools, which appears to have contributed to somewhat lower adoption rates of the most invasive monitoring categories (continuous webcam monitoring, granular keystroke logging) relative to the US market, though comprehensive comparative adoption data remains limited.
05 — What Actually Correlates With Genuine Productivity
The research literature on remote and hybrid work productivity, largely separate from and considerably more rigorous than the vendor-produced case studies that dominate monitoring software marketing, consistently points toward a different set of interventions as more strongly associated with genuine output and retention: clear, outcome-based goal setting (measuring what was actually accomplished rather than how the work looked while being performed), regular but not excessive structured check-ins focused on progress and obstacles rather than activity verification, and organisational trust and autonomy, which multiple studies — including extensive research from Stanford economist Nicholas Bloom on remote work productivity — have found to be more strongly associated with sustained high performance than surveillance-based accountability mechanisms.
This does not mean no accountability mechanism is appropriate — clear expectations, regular communication, and genuine performance management remain legitimate and necessary management functions. It means the specific mechanism of continuous, granular activity surveillance appears, based on the available evidence, to be a poor proxy for the outcomes organisations actually want, while carrying documented costs in employee trust, stress, and retention that a more outcome-focused management approach does not carry to the same degree. This is a pattern this series has now documented in several distinct domains — engagement-optimised social media, activity-based workplace monitoring, screen-time-focused digital wellness — where the easily measurable proxy diverges meaningfully from the harder-to-measure thing that actually matters, and where organisations and platforms optimise for the proxy because it is what the available technology can measure, rather than because it reliably produces the outcome they claim to want.
Tomorrow we are addressing a topic that sits at the frontier of biology and computing simultaneously — DNA data storage, the genuinely remarkable proposal to encode digital information directly into synthetic DNA molecules, and whether it can solve the looming crisis in long-term digital archive storage. See you then.
Switched On is a daily technology series covering the ideas, systems, and arguments shaping the digital world. Opinionated. Witty. Occasionally wrong. Always worth the argument.
