Andy Garcia · freelance dev
FRENES
Back to the blog
Management & Science

141 Organizations Tried a Four-Day Week. Burnout Fell Most for People Who Cut 8 Hours or More.

October 19, 2026 7 min read

On July 21, 2025, Nature Human Behaviour published a study of what happens to workers when their whole organization moves to a four-day week without cutting pay. The authors, from Boston College and University College Dublin, followed 2,896 employees in 141 organizations, mostly in the United States, Canada, the United Kingdom, Ireland, Australia and New Zealand, across six trial cohorts that ran between February 2022 and June 2024. Burnout, job satisfaction, mental health and physical health improved. In 12 control companies, they did not.

If a founder asked me whether to try a four-day week, this is the paper I would put on the table. So I read it closely: the full text (Nature sells it, but the NSF public access repository has a free copy of the accepted manuscript) and the peer review file that Nature publishes alongside. The headline holds up. The fine print changes what you should do with it.

What was actually tested

The organizations were recruited by 4 Day Week Global, a non-governmental organization, on a simple rule: 80% of the hours for 100% of the pay. Companies did not have to adopt a literal four-day schedule, but they had to keep pay constant and offer a "meaningful" reduction in time. 82% offered a standard four-day week. Before the six-month trial, 4 Day Week Global ran about eight weeks of webinars to help them redesign work. The paper's example of what goes: "low- and zero-value activities, such as unnecessary meetings."

The hours did move. The company average fell from 39.21 to 34.01 a week, and the most common week went from 40 hours to 32. In the control companies, it went from 38.88 to 39.16.

The outcomes are the employees' own ratings. Burnout fell from 2.83 to 2.38 on a scale of 1 to 5. Job satisfaction rose from 7.07 to 7.59 out of 10. Mental health went from 2.93 to 3.32 and physical health from 3.01 to 3.29, both on 1 to 5. The authors call these small-to-medium effects. In the control companies, burnout went from 2.90 to 2.94 and job satisfaction slipped from 6.66 to 6.47. The gap between the two groups is statistically significant on all four measures.

A 12-month survey found hours still lower (33.91 a week at company level, against 38.97 at baseline) and the gains still there. Only job satisfaction showed signs of drifting back, while staying above where it started.

The finding most summaries skip

For a manager, the useful result is not the average. It is the dose.

Compared with control employees and with people whose hours did not drop, and after adjusting for background factors, those who cut 8 hours or more saw burnout improve by 0.271 points. Those who cut 5 to 7 hours: 0.203. Those who cut 1 to 4 hours: 0.098. That is a clear dose-response, and the top group gained almost three times as much as the bottom one.

At company level, there is no such gradient. Organizations that cut a lot and organizations that cut a little improved about equally against the controls. The authors' reading is that organizational cuts work largely by changing each person's actual hours.

And not everyone's hours moved. In the analysis sample, 31.44% of respondents reported the same or more hours, and 8.96 points of that are control-group employees. That leaves about 715 of the 2,896 trial employees, roughly one in four, whose hours did not go down. The paper lists who was less likely to cut: women, non-white employees, parents of school-age children, respondents in Australia and New Zealand, people in larger organizations and in civil, social and other services. Supervisors, meanwhile, appear to gain more from reduced hours.

So the policy is not the treatment. The hours are.

The mechanism, and one warning sign

Three channels contribute most to the effect, though they do not explain all of it: self-rated work ability, fewer sleep problems, and less fatigue. Work ability, which asks people to rate their current ability against their lifetime best on 0 to 10, rose from 7.04 to 7.83. Fatigue fell from 2.68 to 2.26.

The warning sign is pace. Perceived job demands, meaning tight deadlines and high speed, rose slightly in trial companies, from 3.51 to 3.59 on 1 to 5, although that rise is not statistically different from the control group's. In the regressions, company-wide cuts predicted higher perceived demands for individuals, while a person's own cut of 8 hours or more predicted lower ones. The authors suggest broader restructuring, "probably introducing new responsibilities and tighter deadlines." Fewer hours can also mean denser hours.

What the design cannot show

It is not randomized, and the companies chose to join. The paper says its findings "probably overestimate the true effect of a 4-day workweek." It argues that senior management made the choice, which should limit the bias in employee outcomes. One reviewer pointed out that in smaller companies, senior managers often consult staff before such decisions.

The control group is small and late. It is 12 US companies that attended an information session and did not join, 285 employees in the analysis, less than 10% of the sample, with an overrepresentation of non-profits and professional or social services. They were surveyed from November 2023 to August 2024, while five of the six trial cohorts started in 2022 (the authors report that results hold when they keep only trial companies surveyed in the same period), and fewer of their employees answered: 58% at baseline and 59% at endpoint, against 80% and 73% in trial companies. The peer review file shows how central that group was. The first version, which did not include it, was rejected in March 2024, and one reviewer pointed to a "glaring absence of a control group." The authors replied: "Since submitting our initial manuscript, we have collected data from 12 control companies." The rejection was reconsidered after their appeal, and the paper was accepted in June 2025.

Everything is self-reported, hours included. Trial employees knew they were in a trial, while control employees were not told the surveys were linked to one. Analysis was not blind and the study was not preregistered. The authors answer the obvious objection, that people might overstate their gains to keep the trial, in two ways. The two surveys are six months apart, which makes earlier answers hard to remember. And there is a telling detail: satisfaction with time rose 2.10 points out of 10, against 0.5 to 0.9 for job, life, household finances and relationships. Their argument: people inflating everything would not draw such clear lines between areas. They also read the 12-month results as a sign that the gains are not merely a Hawthorne effect.

There is no output measure. The paper reports no revenue, sales or delivery data. Work ability is the closest thing, and it is a perception.

There is room for survivor bias. Of 5,245 people who answered the baseline survey, 1,691 did not complete the endpoint one, including 370 who were not sent it because they had left the company, gone on leave or stopped taking part.

And the population is specific: 46% professional services, 34% civil, social and other services, 29% non-profits, 82% of employees fully remote or hybrid, mostly high-income English-speaking countries. As of October 19, 2026, the Nature page says the data is not public yet and will be by 2027.

What I would do with it

I am a CTO and an indie developer, and I built BidForge for small French businesses that answer public tenders, so I read this paper with very small teams in mind. Here is how I would use it rather than quote it.

Reorganize first, and write it down. Trial companies generally redesigned work before the trial, so the paper cannot fully separate its effect from the day off. Treat the list of meetings, reports and approvals you remove as part of the treatment, not as preparation.

Measure hours per person, not the policy. The paper asks two simple questions: "How many hours per week do you usually work in your job?" before, and the same question over the last four weeks after. Look at the distribution, not the mean, and look for who stays at five days.

Choose your output measures before day one, because the study did not have any. Two or three numbers you already track, and a decision rule written in advance.

Watch the pace. Add a question on deadlines and speed. If demands rise while hours fall, you compressed the week instead of reducing it.

In France, check what you are copying. The legal full-time week is 35 hours, and the daily maximum is 10 hours, barring exemptions. A four-day week at 35 hours means 8 hours 45 minutes a day: that is a compressed week, which this study did not test. The same 20% cut applied to 35 hours gives 28.

And if you work alone, this paper says nothing directly about you. It studied employees inside organizations.

The four-day week in this study is not a calendar change. It is a cut in low-value work, then a cut in hours, measured person by person. Copy all three, or do not cite the paper.

Sources

A project like this one?

I design and deploy products like this. Let's talk.

Let's talk