Overtime in Japan: 13 hours a month, or 21?

"Our people barely do overtime, maybe ten-odd hours a month." I've heard this from countless executives in Japan, and it's usually not a lie. The official figure agrees. In 2025, full-time workers averaged 13.2 hours of overtime a month.

But ask the same full-time workers how much overtime they did last month and the answer is 20.6 hours, or about 21. The union federation Rengo got 17.7, and nearly a third of respondents said they do overtime that never reaches the time sheet, averaging 16.9 hours a month.

One thing, three numbers. I'm using it to practice one habit. When a claim doesn't match the evidence, first look at who reported it and who's in the denominator.

Where the three numbers come from

The Monthly Labour Survey is sent by the labor ministry to over thirty thousand workplaces every month, and employers fill in what their attendance system says. For 2025 it gives 9.8 hours across all workers, 13.2 for full-time workers (part-timers excluded), 2.3 for part-timers.

Persol's doda surveys fifteen thousand full-time employees aged 20 to 59 each year and asks them to write down last month's overtime from memory. For April to June 2025 the answer was 20.6 hours.

Rengo asked a thousand regular, contract and dispatch workers in 2024 and got 17.7 hours. Then it asked a follow-up. Did you do any overtime that wasn't recorded as working time? 28.4% said yes, and for them it averaged 16.9 unrecorded hours a month.

Where the gap comes from

First, the denominator. The 9.8-hour figure has part-timers in it, about a third of the base, and part-timers do almost no overtime, so the average sinks. Remove them and it's 13.2. Doda only asks full-timers, so 20.6 and 13.2 are the pair worth comparing.

Second, the reporter. 13.2 comes from employers' attendance systems. 20.6 comes from workers' memories. What I've seen inside Japanese companies is simple. A lot of people clock out and keep working. The system records up to the clock-out. The person remembers up to leaving the building. Rengo's figure puts that in statistical form, with a third of people doing unrecorded work, 16.9 hours on average. Add it back and the gap between 13.2 and 20.6 nearly closes.

Third, definitions. In the attendance system, overtime is time after the clock. In the worker's mind it may include work taken home, the half hour of arriving early, the phone call during lunch. Rengo asked about this. Does changing into a uniform count as working time? Half said no. Morning meetings and setup? A quarter said no.

The same gap shows up in bigger numbers. The Monthly Labour Survey implies a 31.4-hour working week in Japan (part-timers included). The Labour Force Survey, which asks individuals, gives 36.8. Employer-reported versus self-reported, five hours a week apart.

None of the numbers is wrong

All three are facts. They answer different questions. 13.2 answers "how much overtime did companies pay for?" 20.6 answers "how much extra do full-time workers feel they put in?" 16.9 answers "how long is the stretch in between that nobody pays for?"

I wrote about the same pattern in the resignation-proxy post, where usage rates ranged from 2% to nearly 50% depending on the denominator and the question. Overtime is the same. When you see a number, ask three things: who reported it, who's in the denominator, and what time window it covers.

But the number itself doesn't matter much

This is where I should offer a method for finding the true overtime figure. I don't have one. Matching attendance records against PC logs, or counting how often exit interviews mention time. These prove that some people keep working after clocking out. They don't tell you how long, and they change nothing.

More to the point, after years of diagnosing organizations I've come to think overtime hours are a number not worth chasing. Thirteen or twenty-one, the eight-hour difference isn't the problem. The problem is what those 160-odd hours produce. A large part of Japanese overtime isn't too much work. It's low output per hour. Too many meetings, slow decisions, waiting for the boss, waiting for a reply, revising the deck a fifth time. Cut a low-output team's overtime from 21 hours to 13 and what you get is a team with lower output.

So when a client tells me overtime is low, I no longer go check the number. I ask what this team produced last month.

Two things for foreign companies in Japan

First, when reporting your Japan entity's overtime to HQ, say which number it is. Report the official 13 hours and HQ decides the Japanese are coasting. Report the self-reported 20 and you're asked why it differs from government data. Report both and explain the gap. That's more credible than either one alone.

Second, when a candidate asks in an interview how much overtime there is, the reference in their head is the 20-hour kind, not the 13-hour kind. Say "13 hours on average" and they'll add seven from their own experience.

The Monthly Labour Survey had a sampling problem exposed in 2019 and the series was revised afterward, so long-run trends need care. The self-reported surveys are all online panels, which skew toward people who care about working hours, so 20.6 may run high. The true figure sits somewhere between 13 and 21. I don't know where, and I'm not planning to find out.

Which kind of number is your company's overtime figure?