898,000 fewer people, and the work did not shrink

The Statistics Bureau's population estimate for 1 October 2024 is short and unpleasant.

Total population 123,802,000, down 550,000 in a year, a fall of 0.44%. Japanese nationals 120,296,000, down 898,000, a fall of 0.74%. People aged 15 to 64, the group that does most of the paid work, 73,728,000, down 224,000, now 59.6% of the country. People aged 65 and over, 29.3%, the highest share ever recorded.

The foreign population rose by 342,000, a net inflow for the third year running. That softened the total and it is why the headline says 550,000 rather than 898,000. It did not change the direction.

Where this lands inside a company

Nobody experiences a national statistic. What people experience is a team of six that used to be eight, a hiring post that stays open for five months, and a retirement that takes twenty years of process knowledge out of the door on a Friday.

Most Japanese companies we meet still describe automation as an efficiency initiative, something to justify with a payback calculation. That framing made sense when there was always another hire. The estimate says there is not. The orders, the invoices and the customers do not shrink because the headcount did.

The mistake we see most often

The instinct under pressure is to automate everything at once, with a big platform and a target of removing whole jobs. In our experience those projects stall in the middle more often than they finish, and while they stall, the six person team is still doing the eight person job.

The version that works is smaller and less impressive in a slide. Pick one process that one team hates. Watch how it is actually done, not how the manual says it is done. Automate only the steps where a machine is clearly better, and leave the judgment calls with the person. Then measure hours saved, tell the team, and pick the next one.

We have not met a company that regretted starting that small. We have met several that regretted starting big.

The part AI changed

Five years ago, much of office work was too messy to automate. It lived in emails, PDFs, phone calls and half filled spreadsheets. That is precisely the kind of work generative AI is now good at: reading, summarising, extracting, drafting, routing. The messy middle of a Japanese office is where the gain is, and it is within reach of a company of thirty people, not only a bank.

Some of it is subsidised. Japan renamed its IT adoption subsidy to include AI this year, and the training subsidy covers a large share of teaching staff to run what gets built.

The clock

Every year the Statistics Bureau publishes its estimate for 1 October, and nothing in the numbers suggests the next one will show more Japanese people of working age than the last. A company can plan around that or be surprised by it. If you want to start, bring us the one process your most stretched team complains about most. That conversation takes an hour and it is where every automation we have built began.