The pilots everyone cites cut about four hours a week, not eight. Iceland, the case held up as proof, never ran a four-day week at all. And the best-instrumented trial in the world found workers healthier, firms no better off, and the pace of work significantly faster.
The four-day week arrives pre-loaded with statistics. Productivity up 40 per cent at Microsoft. Revenue up 35 per cent in the UK pilot. Eighty-six per cent of Iceland already doing it. Ninety-two per cent of companies keeping it.
Every one of those numbers is either measuring something other than what you think, or is not what its own source says. That is worth establishing carefully, because underneath the bad numbers there is a real and rather more interesting finding.
Most trials cut about four hours, not eight
A four-day week, as sold, means 40 hours becoming 32 — the “100:80:100” formula of full pay for 80 per cent of the time. Here is what the trials actually delivered.
The UK pilot of 61 companies moved workers from 38 hours to 34 — four hours, not eight. Fifteen per cent of employees reported working more. The six-country study published in Nature Human Behaviour recorded 39.12 hours falling to 34.48. The German pilot managed 3.95 hours, with only 34 per cent of organisations reaching a full 20 per cent cut and 46 per cent cutting 10 per cent or less.
These are real reductions and they are not nothing. They are also roughly half of what the phrase “four-day week” describes.
Iceland never had a four-day week
This is the claim that has travelled furthest from its source. The Autonomy and Alda report on the Icelandic trials describes moving public sector workers from 40 hours to 35 or 36 — about an hour less per day. Not four days. The report says so plainly.
The famous “86 per cent” line reads, in full, that 86 per cent of Iceland’s working population “either moved to working shorter hours or have gained the right to shorten their working hours.” A right, not a reduction.
What that right amounts to is documented by the Icelandic commercial workers’ union itself. VR’s own member guidance puts the automatic entitlement at “9 mins per day, 45 minutes a week.” Anything beyond that has to be negotiated workplace by workplace, and generally requires workers to trade away paid coffee breaks. A 2025 analysis of 3,617 English-language news articles put the realised reductions at 0 to 65 minutes a week.
The Icelandic trials also cost money. The report estimates increased costs to the Icelandic government of 4.2 billion krónur a year.
Microsoft Japan was five Fridays of paid leave
The “40 per cent productivity boost” comes from a one-month experiment in August 2019 in which about 2,300 staff were given five Fridays of special paid leave. The metric was sales divided by headcount, for that single month, compared with the same month a year earlier.
Microsoft Japan later added a note to its own page stating that the figure, while factual, was “not a result achieved by this challenge alone” but reflected various factors. The company did not adopt a four-day week. That correction almost never appears alongside the statistic.
What the best-run trial actually found
The German pilot is the most rigorously instrumented study in this literature. It put Garmin wearables on 140 participants across 15,110 observation days rather than relying on how people said they felt.
The physiological results are genuinely good: 89 fewer high-stress minutes and 38 more minutes of sleep per week, and 1,848 more steps.
The organisational results are not. Revenue and profit data, from 12 organisations, showed no significant change. Job satisfaction: no significant change. Turnover: none. Sick days: none.
And one finding the movement rarely quotes: the pace of work rose significantly while workload stayed the same. People were not doing less. They were doing the same amount faster.
Two years on, the same team went back and reached 40 of the 41 organisations. Thirty per cent had reverted, citing economic pressure and workload peaks. Every reverting organisation reported worse work–life balance afterwards.
The one randomised trial
Almost everything above shares a fatal design feature: the companies volunteered. Firms with enough slack to try a four-day week are not a random sample of firms.
There is one exception. A Swedish study randomised 33 public sector workplaces — 580 workers — to a 25 per cent cut in hours at full pay, and followed them for 18 months with diary data. Workers slept 23 minutes longer on workdays, reported better sleep quality, less sleepiness and less stress, all at p<0.002.
And the authors report the effect sizes were small (Cohen’s f² < 0.08). Real, replicable, modest.
The other clean identification comes from France, whose 35-hour laws rolled out by firm size. An IMF analysis found that against a four-hour statutory cut, actual hours in large firms fell 20 to 30 minutes relative to small ones. Aggregate employment: no significant effect. Dual job holding rose 3.3 percentage points. And French workers “did not become more satisfied with their hours of work.”
Where output is time, it costs money
Gothenburg ran a six-hour day for about 60 assistant nurses at the Svartedalen care home. Health outcomes improved markedly. It also required hiring 15 new full-time staff, at a cost of 9.8 million kronor over 18 months against 4.7 million in state unemployment savings. The trial ended as scheduled and was not extended.
The strongest UK public sector case, South Cambridgeshire District Council, is more encouraging: of 24 externally analysed indicators, 22 improved or held steady, voluntary turnover fell 39 per cent, and agency costs dropped £483,000. Two indicators got worse — rent collection, and the time taken to re-let housing. The council’s own analysts note the study “cannot prove it was the trial that caused any changes identified.”
What the evidence actually supports
Read together, the primary sources point somewhere more modest and more useful than the headlines:
- Cutting hours reliably improves worker health. This holds in the randomised trial, in the wearable data, and across the Swedish literature.
- It does not reliably improve firm performance. The only trial with objective financial data found nothing.
- The health gains may depend on cutting workload, not just hours. A Swedish review concludes that proportional workload reduction, rather than fewer hours alone, is what drives the benefit — which is precisely what “same output in less time” does not promise.
- Iceland is a real success, misdescribed. A durable public sector reform from 40 hours to 35 or 36, reported worldwide as something it was not — so the actual achievement is invisible behind a claim that is false.
What remains unsettled
Nobody has randomised firms into a four-day week. Whether the benefit comes from the free day or from the process redesign that accompanies it is unknown — in Germany, 65 per cent of firms cut distractions and 52 per cent changed how they ran meetings, so the day off may be a forcing device rather than the active ingredient. Six months is the standard endpoint, and the one two-year follow-up shows meaningful reversion. And almost every positive result comes from small white-collar firms; where output is proportional to staffed hours, the honest finding is that it costs money.
The four-day week may still be a good idea. But it has been argued for with numbers that do not survive being looked up, and the case for it would be stronger without them.
