Is Remote Work More Productive? What the Research Actually Shows

Hybrid work, one to three days at home a week, is roughly neutral for productivity and good for retention. That is the most consistent finding across the randomised trials. Fully or mostly remote work is a different question. For focused, independent, countable tasks it has often raised output. For work that depends on coordination, feedback, or learning on the job, the better studies find real costs, and the people who pay them are mostly juniors.

So "is remote work more productive?" has no single answer. It depends on how many days, which tasks, whose output is being measured, and over how long. This page sets out the major studies side by side: setting, design, sample size, finding, and the caveat that limits each one. Use it to check a number before you repeat it.

Why the studies seem to disagree

Headlines quote these papers as if they answered one question. They answer four.

How remote? "Remote" covers a hybrid schedule with two home days, four or five home days a week by choice, and a whole firm sent home overnight during a pandemic. Mixing them is the most common error in remote work productivity statistics. In this review, hybrid work is a schedule that splits the week between home and office, usually one to three home days. Fully remote work is working from home, or anywhere, on every workday.

Which task? A call-centre agent's output is calls handled. A software engineer's output includes code other people review and the juniors they train. Quiet homes help the first. Distance hurts the second.

Whose productivity, measured when? Output per hour, performance ratings, promotions and code quality can move in different directions. Bloom's 13% Ctrip gain is mostly more minutes worked: about 9 of the 13 points.

Who chose? A selection effect is a gap that comes from who picks an arrangement rather than from the arrangement itself. If less productive workers sort into remote jobs, remote looks worse even if home costs nothing. If eager volunteers sort in, it looks better. Two studies below separate these effects directly, Emanuel and Harrington's call centres and Atkin, Schoar and Shinde's data-entry trial, and both find selection is large. Ctrip points the same way: when staff could re-choose, the gain rose from 13% to 22%.

The evidence at a glance

Randomised trials: the strongest evidence

Randomisation is the only design that cleanly separates what home does to people from who chooses it.

Ctrip call centre (Bloom, Liang, Roberts and Ying, 2015)

• Setting: Ctrip, a NASDAQ-listed Chinese travel agency with about 16,000 employees. Of 994 airfare and hotel booking staff, 503 volunteered and 249 met the eligibility rules (NBER working paper).
• Design: Randomised by birthday, nine months. The treatment group worked from home four days a week and came in on the fifth.
• Finding: Performance rose 13%. About 9 points came from more minutes per shift (fewer breaks and sick days), about 4 from more calls per minute in a quieter environment. Attrition roughly halved. When Ctrip offered the option firm-wide and let people re-choose, the gain rose to 22% (Quarterly Journal of Economics, 2015).
• Caveat: Independent, countable work by volunteers, with little need to coordinate. Promotion rates fell even after controlling for performance. The most quoted number in the field travels least to knowledge work.

Trip.com hybrid trial (Bloom, Han and Liang, 2024)

• Setting: 1,612 university-educated employees in the Airfare and IT divisions of Trip.com: engineers, marketing and finance staff, 395 of them managers (Nature, 2024).
• Design: Randomised trial, August 2021 to January 2022. Employees with odd birthdays could work from home on Wednesday and Friday. The rest came in five days.
• Finding: Quits fell by a third, from 7.20% to 4.80%, with the largest drops among non-managers, women, and long commuters. Null equivalence tests, which check whether any effect is small enough to rule out, found no effect on performance grades over two years of reviews, and no difference in promotions. Managers expected hybrid to cut productivity by 2.6% before the trial. Afterwards they thought it raised productivity by 1.0%.
• Caveat: One firm, one country, and an option rather than a mandate. Trip.com's chairman, James Liang, is a co-author; the paper discloses this and states that no funding came from the firm. The trial was registered after it started but before data were analysed.

BRAC office-days trial (Choudhury, Khanna, Makridis and Schirmann, 2026)

• Setting: 148 professional employees at BRAC, a large NGO in Bangladesh (Harvard CID summary).
• Design: Nine-week randomised field experiment varying how many days a week each person had to be in the office.
• Finding: Employees with an intermediate number of office days, about two a week, reported better work-life balance, higher job satisfaction and less isolation than those with more or fewer days. Their performance ratings were no different (Review of Economics and Statistics, 108(4), 2026).
• Caveat: Small sample, short window, conducted in 2020, and performance measured by manager ratings. An earlier working-paper version reported better ratings for hybrid workers; the published version reports no difference. Quote the published one.

Chennai data entry (Atkin, Schoar and Shinde, 2023)

• Setting: Data-entry workers hired for real eight-week jobs in Chennai, India. 235 workers recruited from 892 applicants starting in January 2017 (NBER working paper, revised December 2023).
• Design: Randomised assignment to home or office, after first asking where each worker preferred to work.
• Finding: Workers assigned to home were 18% less productive. Two-thirds of the gap appeared on day one; the rest came from faster learning in the office. Workers who preferred home were 12% faster and more accurate at baseline, but were 27% less productive at home than in the office, compared with 13% for those who preferred the office. Childcare and home demands explain part of this.
• Caveat: Low-income setting, often crowded homes, new workers on short contracts, and not peer-reviewed.

Natural experiments from real firms

These studies use real company data and a sharp change in policy. They are larger than the trials but cannot fully rule out other things changing at the same time, especially during COVID.

HCL Technologies IT professionals (Gibbs, Mengel and Siemroth, 2023)

• Setting: More than 10,000 skilled professionals at HCL Technologies, an Indian IT services firm, with personnel and Microsoft Workplace Analytics data from April 2019 to August 2020 (Journal of Political Economy Microeconomics, 2023).
• Design: Before and after the firm-wide switch to working from home in March 2020.
• Finding: Hours rose, much of it from starting earlier and finishing later. Output declined slightly. Output per hour fell 8% to 19%. Time in meetings and coordination rose, and uninterrupted focus hours shrank. People networked with fewer colleagues and had fewer one-to-one meetings with supervisors. Employees with children at home saw bigger productivity declines.
• Caveat: Abrupt, fully remote, during a pandemic with Indian schools closed. The authors check that the change lines up with the switch to WFH rather than infection rates, but this is not hybrid and not chosen.

Fortune 500 call centres (Emanuel and Harrington, 2024)

• Setting: Call centres at a US Fortune 500 firm that hired for identical jobs on-site and remotely, with workers hired from July 2018 (American Economic Journal: Applied Economics, 16(4), 2024).
• Design: Compares remote and on-site workers before COVID, then uses office closures to see how much of the gap survives when everyone is remote.
• Finding: Before COVID, remote workers took about 12% fewer calls per hour. When offices closed, the gap narrowed by about 4 points, leaving an 8% gap even with everyone at home. The authors read that remainder as selection: less productive workers chose the remote jobs. Formerly on-site workers saw lower call quality and fewer promotions once remote.
• Caveat: One firm, one job type.

Tempo BPO call centre, Turkey (Aksoy, Bloom, Davis, Marino and Özgüzel, 2025)

• Setting: Tempo BPO, a large Turkish call centre, with data on 1,766 agents from January 2019 to January 2023 (BFI working paper, May 2025).
• Design: Before and after a shift to fully remote work that the firm kept after lockdown ended.
• Finding: Calls per hour rose 9.1% during lockdown and stayed 10.5% above 2019 levels afterwards, with better customer service ratings. Quieter homes were the main driver. Remote hiring increased the share of women and graduates without raising wages. Remote employees who had in-person training at the start had higher long-run productivity and lower attrition.
• Caveat: Working paper, single firm, no randomisation. Again, countable independent tasks.

USPTO patent examiners (Choudhury, Foroughi and Larson, 2021)

• Setting: US Patent and Trademark Office examiners moving from a work-from-home programme, which required living near the office, to work-from-anywhere (Strategic Management Journal, 42(4), 2021).
• Design: Natural experiment. Timing of the switch was set by union-management negotiations.
• Finding: Output rose about 4.4% with no increase in rework.
• Caveat: Experienced examiners doing independent, well-measured work. It compares two kinds of remote work, not remote with office.

What remote work does to collaboration and learning

The output studies above measure individuals. These two measure the work between people, where distance costs show up.

Microsoft network silos (Yang et al., 2021)

• Setting: More than 61,000 Microsoft employees, most of its US workforce (Nature Human Behaviour, 2021; Berkeley Haas summary).
• Design: Firm-wide WFH mandate as a natural experiment, comparing the roughly 18% already remote with those who moved, matched on role, tenure and group. Content-free metadata on email, chat, calls and meetings.
• Finding: Networks became more static and siloed. Time collaborating across business groups fell about 25%. Communication shifted from synchronous to asynchronous. Meeting hours fell about 5%. A colleague's remote status changed a person's own habits too, which matters for hybrid scheduling.
• Caveat: Measures collaboration patterns, not output. Single firm, during a pandemic, with several Microsoft-employed authors. The productivity implications are the authors' inference.

Proximity and code feedback (Emanuel, Harrington and Pallais, 2026)

• Setting: Software engineers at a Fortune 500 firm from 2019 to 2024 (NBER working paper 31880, revised June 2026).
• Design: Difference-in-differences, which compares how two groups change after the same event, using two shocks: the 2020 office closures and the 2022-23 return-to-office mandates, comparing co-located teams with already-distributed ones.
• Finding: Sitting near teammates increased coding feedback by 18.3% and improved code quality. The gains went to less-tenured and younger engineers. Experienced engineers wrote less code when sitting near teammates: mentoring has a cost. In national US data, young graduates in remotable jobs saw unemployment rise relative to older graduates after the pandemic, a pattern absent in non-remotable jobs.
• Caveat: Working paper, one firm for the main results, and the national finding is correlational.

Surveys and macro data: useful, but not proof

Surveys tell you what people believe, not what caused what. Read them as context.

• How much remote work there is: About 26% of paid US workdays were worked from home in June 2026. Among full-time employees from July 2025 to June 2026, 12% were fully remote, 26% hybrid, and 62% fully on site (SWAA, Barrero, Bloom, Buckman and Davis, July 2026 update; monthly online survey, over 200,000 observations).
• Engagement by location: In May 2026, 37% of exclusively remote US employees were engaged, 34% of hybrid, and 29% of on-site employees in remote-capable jobs (Gallup). Engagement is not output. Among remote-capable employees, 52% are hybrid, 26% exclusively remote and 22% on site.
• What managers say: 69% of managers said hybrid or remote work made their team more productive, 12% less, and 19% saw no change. This is a vendor survey of 2,000 US full-time workers, 72% of them managers, and Owl Labs sells hybrid-meeting hardware (Owl Labs, July 2025).
• The trust gap: In 2022, 87% of employees said they were productive. Only 12% of leaders had full confidence their team was, and 85% said hybrid made that confidence hard. This is vendor research: an Edelman survey of 20,006 knowledge workers in 11 countries for Microsoft (Work Trend Index).
• The macro view: A Federal Reserve note finds that US sectors with a larger share of remote-ready jobs show no immediate productivity change. They do show higher labour productivity after about three years (Tito, FEDS Notes, August 2025). It measures the ability to work remotely, not remote work itself, and the author flags that it may partly capture broader technology adoption.

The Work Trend Index numbers are the gap the research has not closed. Employees and their managers are looking at different evidence. That is a visibility problem as much as a productivity one, which is the argument of remote trust is a signal problem.

What the evidence supports

Taking the studies together, five conclusions hold up.

1. Hybrid is neutral to slightly positive for productivity, and clearly positive for retention. The two hybrid trials found no performance penalty: Trip.com with two optional home days a week, BRAC with about two office days a week. Trip.com found a one-third drop in quits. No randomised study we found shows hybrid lowering measured performance.
2. Fully or mostly remote work depends on the task. Independent, measurable, concentration-heavy work (calls, patent examination) has gained 4% to 13%. Work that needs constant coordination lost ground in the HCL data. Some of the difference is the pandemic. Some is the work.
3. The persistent cost of distance is learning. Juniors get less feedback and mentoring when remote. Seniors get more of their own work done. Firms bear the cost later, in skills, which is why it rarely shows up in a six-month productivity number.
4. Selection is large. Comparing remote and office workers without a design that handles who chose what can produce almost any answer. Treat raw comparisons with suspicion, including flattering ones.
5. Belief lags evidence. Trip.com's managers moved from expecting a 2.6% loss to seeing a 1.0% gain once they had run the experiment. Experience of the trial, not argument, changed their minds.

The evidence supports neither "remote workers are more productive" nor "remote work kills productivity" as a general claim.

What this means if you are asking for remote days

If you are using this research to ask for home days, it points to a specific ask.

Ask for hybrid, not fully remote. Two mid-week home days is the design that tested cleanly at Trip.com, and BRAC's best results came with about two office days a week. Fully remote is a minority arrangement (12% of full-time US employees in SWAA) and the research on it is mixed.

Cite the right study. Lead with Trip.com, not Ctrip's 13%. The call-centre result does not describe your job. Gibbs, Mengel and Siemroth tell you what to promise: coordination rules, protected focus time, regular one-to-ones.

Address learning directly. If you mentor juniors, say which office days you will keep for it. If you are junior, the proximity evidence suggests you may lose more than you gain from distance early on. Trip.com excluded interns and staff on probation from its trial for that reason.

Make output visible. The trust gap is a visibility gap. A manager who cannot see you work needs another way to see what you did. A written trail of what you shipped on home days gives them that. That is the problem DoneThat, which publishes this blog, is built for: automatic summaries of your work that you control and choose to share, rather than activity monitoring. A shared doc of weekly progress does the same job if you keep it up.

For the full playbook, read how to convince your boss to let you work remotely, then use the remote work proposal template. If your company is moving the other way, see how to respond to a return-to-office mandate and the research on proximity bias. For the cost side of the ledger, see the true cost of commuting. Everything in this series is collected under remote work.

Frequently Asked Questions

Is remote work more productive than working in the office?

Not in general. Hybrid work shows no measurable performance penalty in randomised trials. Fully or mostly remote work has raised output for independent tasks like call handling and lowered it for coordination-heavy IT work during lockdown. The answer depends on the number of home days and the type of work.

What is the best study on remote work productivity?

For hybrid knowledge work, the Trip.com randomised trial of 1,612 employees (Nature, 2024) is the strongest. For mostly remote independent work, the Ctrip call-centre trial (2015), where staff worked from home four days a week. For why remote comparisons mislead, Emanuel and Harrington (2024).

Does remote work hurt careers?

The evidence is mixed by design. Working from home four days a week lowered promotion rates at Ctrip, and full-time remote work lowered them among formerly on-site call-centre workers. Two optional home days at Trip.com did not affect promotions over two years. Junior staff appear to lose the most mentoring from distance.

How to cite this page

Don (DoneThat). "Is Remote Work More Productive? What the Research Actually Shows." DoneThat Blog, 3 November 2026. https://donethat.ai/blog/remote-vs-office-productivity-research

You are welcome to reproduce the summary table with a link back to this page. For any individual figure, please cite the original study linked next to it.

Written by Don, DoneThat's AI coach, so this article is AI-written. Every study was read at its source, and every figure is linked to the paper or publisher page it comes from. Working papers and vendor surveys are labelled as such. Reviewed by Christoph Hartmann, DoneThat's founder. Read more about DoneThat, or send a correction through support.