Randomised trials find that two days a week at home costs nothing in measured output and cuts quitting by about a third. The largest study of corporate mandates found no gain in profit or firm value. But the evidence does show one real cost of working apart: junior staff learn less.
The retreat from pandemic-era remote work ended in 2023. In the Global Survey of Working Arrangements, a stratified survey of college-educated employees, the balanced panel of 23 countries surveyed in every wave averaged 1.55 paid days a week at home in 2022, 1.29 in 2023 and 1.23 in late 2024 and early 2025.7 The first step down was a fall of 16.8%; the second was 4.7%. Roughly a quarter of all workdays are now done at home.
Britain sits at the high end. Among the countries in the survey’s latest wave, college-educated employees in the English-speaking economies reported roughly 1.5 to 1.9 days a week at home, against well under one day in several advanced East Asian economies.7 The Office for National Statistics puts 28% of working adults in Great Britain in a hybrid pattern between January and March 2025, a share that has risen gradually since March 2022 while the proportion travelling to work only has fallen.8
Hybrid working is not evenly distributed, and neither is the return-to-office fight. ONS survey data for January to March 2025 shows 41% of workers with a degree or equivalent hybrid working, against 4% of those with no qualifications — a ratio of 10.3 to one. By income the gap is 45% for those earning £50,000 or more against 8% for those under £20,000, a ratio of 5.6 to one.8
The pattern is consistent across cuts: hybrid working is more common among full-time workers than part-time (34% against 18%), among employees than the self-employed (30% against 24%), and among those living in the least deprived fifth of England (32%) than the most deprived (24%).8 Roughly half the UK workforce holds a job that cannot be done remotely at all. Whatever is decided about office attendance applies to a minority, and a comparatively advantaged one.
UK employers have tightened attendance rules without abandoning hybrid working. In the CIPD’s 2025 survey, 74% of organisations had hybrid working in place, down from 84% in a comparable 2023 survey; 65% required a minimum number of days on site, up from 52% in 2023. Among those setting a weekly minimum, three days was the most common requirement, at 48%. A further 14% of employers allowing hybrid working planned to introduce or increase mandatory days within a year.13
More than half of employees (53%) reported a great deal or a fair amount of pressure to spend more time in the workplace, and 70% of those said it came from senior leaders.13 What is striking is the reason employers give for that pressure.
The three leading answers are relational: connections and relationships (55%), collaboration (54%) and engagement (47%). A direct productivity claim — that people do not work as effectively at home — is given by 26%, behind monitoring performance (27%). Onboarding new joiners (34%) and training and development (33%) sit in the middle of the list. On the employers’ own account, then, the mandate is mostly an argument about cohesion, not about output.
There is no single number for the productivity effect of working from home, because the studies measure different outputs in different jobs under different arrangements. Exhibit 04 places the best-identified estimates on one index, with office-based output set at 100.
The two positive call-centre results share a mechanism. At Ctrip in 2010–11, home working raised performance 13%, of which 9 points came from working more minutes per shift and 4 points from more calls per minute, attributed to a quieter environment; attrition halved, but promotion rates conditional on performance fell.2 At a Turkish call centre that went fully remote, productivity rose about 10% through shorter calls, again attributed to a quieter home setting; the firm also raised its share of graduate employees by 14% without raising wages, by reaching workers who had been outside the local labour pool.10 A third positive finding, a 4.4% rise in output among US patent examiners, compares working from anywhere with working from home rather than with the office, and so does not belong on the same index.6
The negative results come from settings where remote work was imposed and total. At a US Fortune 500 firm, closing the offices cut formerly on-site workers’ calls per hour by 4% relative to already-remote colleagues, and reduced call quality and promotion rates; most of the raw 12% gap between remote and on-site staff turned out to be selection rather than treatment.3 Among more than 10,000 skilled professionals at an Indian technology company, hours worked rose while output fell slightly, so productivity fell by 8–19%; the authors attribute much of it to higher communication and coordination costs.4
One reading — and it is a reading, not a measurement — is that the sign of the effect tracks how much of the week is spent apart and how much the work depends on other people. The only randomised test of the arrangement most firms actually operate is the Trip.com trial: 1,612 graduate employees in engineering, marketing, accounting and finance, randomly assigned by birth date either to five days in the office or to three in and two at home for six months. Formal equivalence tests found no effect on performance grades over the next two years of reviews, promotion rates were unaffected, job satisfaction rose and quit rates fell by about a third, with the retention gain concentrated among non-managers, women and those with long commutes.1
Two working papers from the University of Pittsburgh provide the only large-sample evidence on the mandates themselves. Neither is peer-reviewed, and both should be read as strong description rather than settled causal fact.
The first identified S&P 500 firms that had announced return-to-office mandates and tested what followed. Mandates were more likely at firms whose share prices had recently fallen and at firms with male, powerful chief executives. Difference-in-differences estimates found significant declines in employees’ Glassdoor ratings of job satisfaction after a mandate, but no significant change in financial performance or firm value. The authors read this as inconsistent with the argument that managers impose mandates because they expect them to raise firm value.11
The second tracked the employment histories of more than three million workers at 54 S&P 500 technology and financial firms that announced mandates between the second quarter of 2022 and the second quarter of 2023, using LinkedIn profile data supplied by Revelio Labs. Abnormal turnover — each firm’s rate less the national average — rose after the announcements, and it rose unevenly.12
Turnover rose about 14% overall. The increase was statistically significant for women (21.2%), mid-level managers (18.2%), top managers (21.5%) and the most-skilled quartile (18.6%), and not significant for men, rank-and-file staff or less-skilled employees. Replacing those people took longer: the average time to fill a vacancy rose by roughly 23%, about 12 days on a 51-day baseline, and hire rates fell.12 Because large firms usually have below-average turnover, the authors argue the sample biases against finding an effect; they also state plainly that they cannot draw causal inferences, though parallel-trend and placebo tests are consistent with the timing.
The best-identified benefit of physical co-presence is not output. It is training. Economists at the Federal Reserve Bank of New York, the University of Iowa and Harvard studied software engineers at a Fortune 500 firm from 2019 to 2024, using two shocks to proximity — the 2020 office closures and the firm’s own 2022 and 2023 return-to-office mandates — which affected co-located teams more than teams already split across buildings. Sitting near teammates raised the volume of code-review feedback by 18.3% and improved code quality, with gains concentrated among less-tenured and younger engineers. The trade-off is direct: experienced engineers wrote less code when sitting near colleagues, because they were spending time on other people.9
A separate study of 61,182 US Microsoft employees over the first six months of 2020 found that firm-wide remote work made collaboration networks more static and siloed, cut the share of collaboration time spent with contacts outside a worker’s own group by about 25%, and shifted communication from synchronous to asynchronous channels. The authors note these effects may make it harder to acquire and share new information across an organisation.5
This is a serious case, and it is the one the evidence supports most cleanly. It is also, on its own terms, an argument for co-presence at particular moments — induction, review, cross-team work, the first years of a career — rather than for a uniform five-day rule applied to everyone. Both findings come from periods of total remote working, not from a two-day week at home.
Asked whether increased home or hybrid working has changed their organisation’s productivity and efficiency, UK employers have given a stable answer since 2021: rather more say it has risen than say it has fallen. The net balance was +10 in 2020, +23 in 2021, and +25 in both 2023 and 2025.13
Employers also rated the wider effects. The most positive net scores were for attraction and retention of talent (+62), the ability to recruit from a wider geography in the UK (+60), employee financial wellbeing (+56) and the environment (+52). The negative ones were narrower and specific: the ability of managers to lead teams effectively (−13), employees’ connection to organisational purpose (−10) and organisational culture (−2).13 That is close to the pattern in the academic literature — gains in recruitment and retention, costs in coordination and development.
No randomised trial of a return-to-office mandate exists, and none is likely: firms do not assign policies by lottery. The two mandate studies are observational, cover only large US technology and financial firms in 2020–23, and rest on self-maintained LinkedIn profiles and voluntary Glassdoor reviews, both of which are self-selected samples. Neither is peer-reviewed. The randomised evidence on hybrid working comes from single firms — two of them Chinese travel companies — and mostly from measurable, individually-assessed roles; there is no comparable trial in a UK organisation. Nothing in the literature measures the long-run effect of any arrangement on innovation, and the training effects identified for junior engineers have not been priced against the retention effects identified for senior staff. Finally, most of the negative productivity findings come from 2020–21, when home working was compulsory, schools were shut and firms had not adapted their processes.
Two questions are being argued at once and they are not the same kind of question. The first is empirical and evidence can move it. The second is about who bears a cost, and evidence cannot settle it.
Working apart has measurable costs that fall on the young and on the organisation rather than on the individual. Proximity raises code-review feedback by 18.3% and improves code quality, with gains concentrated among less-tenured engineers.9 Firm-wide remote work cut cross-group collaboration time by about a quarter.5 Where remote work was total and the job depended on coordination, productivity fell by 8–19%.4
The costs are real but the remedy is not. The only randomised test of a two-day-at-home schedule found no effect on performance grades or promotion and a fall of about a third in quitting.1 Across the S&P 500, mandates were followed by lower job satisfaction and no significant change in financial performance or firm value11, and by higher turnover among exactly the senior and skilled staff hardest to replace.12
An employer that pays for a workplace may reasonably require people to use it, and uniform rules are fairer to the roughly half of the workforce whose jobs cannot be done from home. In the CIPD survey, 25% of employers giving reasons for more attendance cited fairness between office- and non-office-based roles.13
A mandate transfers time and money from employee to employer without compensation, and it does so unevenly: the retention effects fall hardest on women and long-distance commuters.112 Around 1.1 million UK employees say they left a job in a year because of a lack of flexible working.13
Where the evidence points. On the narrow empirical claim — that requiring more office attendance improves business performance — no study has found it, and the largest study designed to detect it found no significant change in financial performance or firm value.11 That is an absence of demonstrated benefit rather than proof of no benefit. On the value questions, this desk takes no position.
Every estimate used above, with its design and the outcome it actually measures.
| Study | Setting | Design | Sample | Headline result |
|---|---|---|---|---|
| Bloom, Han & Liang (2024) | Trip.com, Shanghai | Randomised trial | 1,612 | Two days at home: no effect on performance grades; quits down about a third |
| Bloom et al. (2015) | Ctrip call centre | Randomised trial | 249 | Fully remote four days: performance +13%; attrition halved; promotion fell |
| Emanuel & Harrington (2024) | US Fortune 500 call centre | Difference-in-differences | — | Going remote: calls per hour −4%; quality and promotion fell |
| Gibbs, Mengel & Siemroth (2023) | Indian IT services firm | Personnel and analytics data | 10,000+ | Forced remote: hours up, output slightly down, productivity −8% to −19% |
| Aksoy et al. (2025) | Turkish call centre | Administrative data | — | Fully remote: productivity +10%; graduate share +14% at unchanged wages |
| Choudhury et al. (2021) | US Patent and Trademark Office | Natural experiment | — | Work-from-anywhere versus work-from-home: output +4.4%, rework unchanged |
| Emanuel, Harrington & Pallais (2026) | Fortune 500 software teams | Difference-in-differences | — | Sitting together: code feedback +18.3%; senior engineers write less code |
| Yang et al. (2022) | Microsoft, United States | Natural experiment | 61,182 | Firm-wide remote: cross-group collaboration time −25%; networks siloed |
| Ding & Ma (2024) | S&P 500 firms | Difference-in-differences | — | Mandates: job satisfaction fell; no significant change in performance or value |
| Ding, Jin, Ma, Xing & Yang (2024) | 54 S&P 500 tech and finance firms | Difference-in-differences | 3m+ | Mandates: abnormal turnover +14%; time to hire +23%; hire rate down |
Material deliberately left out to hold length.