Microsoft Japan Four-Day Week: Five Fridays Off and a 40% Headline

The room

The psychology: Parkinson’s law

C. Northcote Parkinson, a British historian, set out the idea in a satirical essay in The Economist in 1955, drawing on his experience of bureaucracy as an army staff officer. His target was offices where tasks, meetings and paperwork grow to fill whatever hours are on offer. Later researchers turned the joke into testable questions: do tighter limits change how people work, and when do they help or hurt?

The Microsoft Japan trial was built on the hopeful half of Parkinson’s law. If an hour-long meeting fills an hour because it is booked for an hour, then a 30-minute default should shrink it. If a week of work fills five days partly because five days are available, then four days with sharper rules might hold most of the output. The meeting data fits the first part: the share of 30-minute meetings rose by almost half once the default changed.

The research also shows the limits of the idea. Deadlines help most when someone sets them well and spaces them sensibly, and long hours lose value gradually rather than at a sharp cliff. A time limit is a design tool, not a guarantee. It works on work that was padded, and it does nothing for work that genuinely needs the time, such as covering clients who are open on Fridays.

Why smart people fall for it

Three forces explain why a hard time limit changed habits, and why its headline number still needed care.

1. 1

   A fixed limit forces choices

   With one day removed for everyone at once, teams had to decide which meetings and tasks mattered, rather than fitting everything in as before.

2. 2

   Defaults do the daily work

   A 30-minute, five-person meeting default changed thousands of small decisions without asking anyone to try harder.

3. 3

   Metrics shape the story

   Sales per employee in one month moves for many reasons, so a striking figure invited a stronger conclusion than the design could support.

Did the Microsoft Japan four-day week really raise productivity?

The honest answer is that the trial did not show it, and the company itself said so. The 39.9 percent figure compared sales per employee in August 2019 with August 2018. That measure does not count hours at all. It rises if sales grow or headcount falls, and in a software business a month’s sales often reflect deals agreed long before. When Microsoft Japan added its November notice, it was acknowledging exactly this: the number was real, but it was not evidence that five Fridays off had caused it.

The design also made attribution hard. It was a single month compared with the same month a year earlier, with no comparison group. The days off arrived together with new meeting defaults, a push towards chat over email, focus-time tools and boosted benefits, so even a clean result could not say which change mattered. The electricity and printing savings were measured against August 2016 and partly reflect offices simply being closed for five days.

The employee survey is more informative, but it is still the company measuring its own programme. Most responses were positive, and the results briefing added nuance the headline missed: of 856 free-text comments, about one in ten was strongly negative, and the presenter said the result should not be read as a simple nine-to-one vote. Sales staff and consultants, whose work depends on client hours, found the fixed day hardest.

Outside commentary pointed in the same direction. Unseen Japan, in a piece whose author disclosed that he worked for Microsoft, noted that similar schemes had already been tried in Japan, including Fast Retailing’s four ten-hour days, which did not reduce total hours. Fortune quoted the sociologist Phyllis Moen, who argued that cutting office days does little to relieve strain unless workload and constant availability are also addressed. None of this means the trial failed. It means the evidence it produced was about habits and satisfaction, not about output.

What does research say about time limits and output?

The most careful evidence on deadlines comes from Dan Ariely and Klaus Wertenbroch. In classroom and laboratory studies, people were willing to bind themselves with costly deadlines, which suggests they knew they would otherwise put work off. But the self-chosen deadlines were not set well. In the proofreading experiment, evenly spaced deadlines set by the researchers beat both self-chosen deadlines and a single final deadline, on errors found and on money earned.

For managers that is a useful finding. A limit set by the organisation, evenly applied, can work better than leaving each person to manage their own time. Microsoft Japan’s company-wide closure and meeting defaults were limits of that kind. They did not ask each employee to decide how short a meeting should be; they changed the starting point for everyone.

On hours themselves, John Pencavel’s analysis of First World War munition workers found that output kept pace with hours up to roughly 49 a week and then grew more slowly, so that 70 hours produced little more than 56. Weeks without a rest day also produced less. Pencavel was clear that the sample was small, the work was physical and the conditions were unusual, so the numbers do not transfer directly to an office in Tokyo. The broader point does: beyond some level, extra hours add less than they appear to.

Taken together, the research supports a modest version of Parkinson’s law. Well-designed limits can squeeze out padding, and very long hours tend to buy less than their cost. It does not support the idea that any cut in time produces a matching gain in output, and it says nothing about work that is driven by when customers need you.

How would a modern team design the trial?

A modern team would start by choosing the outcome measures before the trial begins, and pick ones the change could plausibly move: hours worked, meeting time, project delivery against plan, customer response times, staff survey scores and voluntary leave taken. Sales per employee can be tracked, but it should be labelled as context rather than a result.

It would build in a comparison. That could be a staggered rollout, with some departments starting a quarter later, or a matched team that keeps its old pattern for a while. Running the trial for several months, across busy and quiet periods, would separate the effect of the policy from the effect of the season.

It would design for the roles that find a fixed day hardest. Microsoft Japan already let some customer-facing staff take their days later, and it later dropped the paid Fridays in favour of flexible leave plus the meeting rules. A modern design might start there: rotating days off for client-facing teams, shared cover, and a clear message to customers about who answers on which day.

Finally, it would cut work, not just time. The parts of the Microsoft Japan programme that lasted were the meeting defaults, the shift away from email and the daily recording of hours with alerts to supervisors. Those are the changes that make a shorter week possible. Without them, a day off can simply push the same load into four longer days.

The manager's playbook

1. Choose your measures before the trialAgree in advance which outcomes count as success, such as hours, delivery and customer response times, so a striking side number cannot become the headline.
2. Keep a comparison groupStagger the start between departments or keep a matched team on the old pattern, so you can separate the policy’s effect from everything else.
3. Change defaults, not just daysSet shorter default meeting lengths, smaller invite lists and clear channels for quick decisions, so the remaining days can absorb the work.
4. Plan cover for client-facing rolesGive sales, support and consulting teams rotating days or shared cover, and tell customers in advance who is available when.
5. Run it across more than one seasonA single month can be distorted by holidays and deal timing; several months show whether results hold.

Questions people ask