The Digital Whale

Work

Technology at work and what it measures

The measured effects of workplace technology are bimodal and setting-dependent, and almost every circulating statistic about tool use comes from vendor telemetry.

The best evidence on workplace technology is narrow, old and contradictory, and it does not support any general claim that tools make organisations more productive. Monitoring raises output sharply where the monitored task is the output — theft in restaurants, calls in call centres — and lowers it where workers need room to experiment. Open-plan offices, the physical counterpart of always-on tooling, cut face-to-face interaction by around 70% in the only study with sensor data. Remote work measurably changed how much people communicate and when, but every such study measures inputs. There is essentially no credible causal evidence that adopting a new collaboration tool raises what a firm produces, and no national statistics office measures how many employers monitor staff at all. For a concrete commercial example of this category, employee monitoring software shows how workplace activity measurement is packaged for organizations.

Monitoring works where the monitored task is the output

The strongest pro-monitoring result comes from restaurants, where revenue and drink-level theft are both directly observable and adoption was staggered across locations.

2015

Installing IT-based theft monitoring reduced employee theft and raised productivity, with the effect coming from changed behaviour among incumbent workers rather than from turnover.

Direction: Increase. Strength of evidence: Strong.

Pierce, Snow & McAfee, Management Science, 2015392 restaurant locations across five firms, staggered IT monitoring adoption

Caveat Service work with trivially measurable output, which is precisely the condition that does not hold in knowledge work.

That last point is the whole argument. The result is clean because a stolen drink is countable. Nothing in it transfers to a setting where the valuable part of the job is judgement.

The same instrument reduces output where discretion matters

The counter-result is a field experiment in a Chinese mobile phone factory, and it runs directly against the intuition that more visibility produces more compliance and therefore more output.

2012

Giving production lines a modest curtain of privacy from managerial observation raised line performance, an effect the author named the transparency paradox.

Direction: Decrease. Strength of evidence: Mixed.

Bernstein, Administrative Science Quarterly, 2012Field experiment on production lines in one Chinese mobile phone factory

Caveat A single factory with a small number of lines measured over weeks, so the direction is well identified and the magnitude is not.

Workers who are watched stop deviating from the official process, including the deviations that were making the process work. Put the two studies together and the honest summary is that surveillance has no single effect: it depends on whether the thing being watched is the thing being produced.

Open-plan offices reduced face-to-face contact

The most-cited workplace-design finding is real but far thinner than its reputation. Two firms moved their headquarters to open layouts and researchers put sociometric badges on employees before and after.

2018

Moving to open-plan layouts cut face-to-face interaction by roughly 70%, with a compensating rise in email and instant messaging.

Direction: Decrease. Strength of evidence: Mixed.

Bernstein & Turban, Philosophical Transactions of the Royal Society B, 2018Two corporate headquarters, wearable-sensor samples of dozens of participants, short pre and post windows

Caveat Two organisations and small samples, routinely over-cited as a settled law about open offices.

Remote work changed communication volume, which is not output

The largest measurement of what the 2020 shift did to working patterns came from email and calendar metadata rather than surveys, which makes it behavioural but also proprietary.

2020

The shift to remote work raised meetings per person by 12.9% and attendees per meeting by 13.5% while cutting average meeting length by 20.1%, a net fall in meeting time of 11.5%, and lengthened the average workday span by 8.2%, about 48.5 minutes.

Direction: Cuts both ways. Strength of evidence: Strong.

DeFilippis, Impink, Singell, Polzer & Sadun, NBER Working Paper 27612, 20203,143,270 users across 16 metro areas, email and calendar metadata

Caveat One vendor's corporate clients during the extraordinary spring 2020 shock, and workday span is first-to-last email rather than hours worked.

That extra 48.5 minutes of span is the measurable root of work that does not switch off. What the metadata cannot say is whether any of it produced more.

The other large study of the same period comes from inside a company with a direct commercial stake in the answer.

2022

Firm-wide remote work made collaboration networks more static and siloed, with fewer bridging ties and a shift from synchronous to asynchronous communication.

Direction: Decrease. Strength of evidence: Mixed.

Yang, Holtz, Jaffe, Suri, Sinha, Weston, Joyce, Shah, Sherman, Hecht & Teevan, Nature Human Behaviour, 202261,182 Microsoft employees, firm telemetry, 2020

Caveat Microsoft researchers using Microsoft telemetry on Microsoft employees, and network structure is a proxy for innovation rather than a measure of it.

Microsoft's annual Work Trend Index, the most quoted source of workplace-technology statistics in the business press, carries the same problem in stronger form. The 2026 edition was published on 5 May 2026 (Microsoft, 2026). Microsoft sells the collaboration and AI products the report measures, its telemetry covers only Microsoft 365 customers, and the landing page discloses no sample or methodology.

Interruptions cost stress rather than time

The popular account of interruptions is that they destroy productivity. The controlled experiment usually invoked in support of it found something more specific and less convenient.

2008

Interrupted work was completed in less time with no loss of quality, but with significantly higher stress, frustration, time pressure and effort.

Direction: Cuts both ways. Strength of evidence: Mixed.

Mark, Gudith & Klocke, ACM CHI, 200848 participants, controlled lab experiment on simulated email tasks

Caveat 48 mostly German student participants on short simulated tasks, which should not be extrapolated to long-horizon creative work.

People compensate for interruption by working faster, and pay for it in strain. That is a different claim from the one made on the slide decks, and it is examined further in what the numbers on focus and interruption actually show.

The short version

  • IT theft monitoring across 392 restaurant locations raised productivity by changing incumbent workers' behaviour, in a setting where output was directly countable (Pierce and colleagues, 2015).
  • A field experiment in a phone factory found that shielding production lines from observation raised their performance, the opposite sign from the restaurant result (Bernstein, 2012).
  • The open-plan finding of a roughly 70% fall in face-to-face contact rests on two organisations and a few dozen sensor-wearing participants.
  • Metadata from more than three million users showed meetings became more frequent and shorter after the 2020 shift, with the workday span lengthening by about 48.5 minutes.
  • The evidence has a hard limit: no study credibly links adopting a workplace tool to firm output, and the widely quoted 23-minute interruption recovery figure has no basis as a constant.