Japan Ran AI Over Its Own Statute Book and Found 10,000 Procedures Nobody Had Counted

In July 2025 the Digital Agency published the results of a survey it is legally required to run, counting every administrative procedure the Japanese state operates. It is not a document anyone was supposed to find interesting. I have read it three times.

The first thing in it is a counting problem. The previous survey, in FY2021, found 61,390 procedure types. This one found 75,085. The agency says plainly how the extra ones turned up: it ran generative AI over the legal data and detected roughly ten thousand additional procedures.

Sit with that. Not ten thousand new procedures created in four years. Ten thousand that already existed, written into law, and were not on the list.

If you have ever inherited a product nobody fully documented, you know this feeling. The first honest thing you do is count what is actually there, and the count is always worse than the handover said.

Half of them are online now, which sounds like progress

The headline numbers are genuinely better than they were. The share of procedure types you can complete online went from about 32% to 51.6%. Among procedures that can be done online, the share of actual filings that were done online went from about 65% to 79.4%.

Four years, twenty points on both measures. On any dashboard that is a good year.

Then you get to page four and the denominator falls apart.

217 procedures are the entire country

Of the 75,085 procedure types, 217 have an annual volume above one million. That is 0.3% of the list. Those 217 account for 95.6% of every administrative transaction that happens in Japan in a year, out of about 4.3 billion.

At the other end, 55,343 procedure types, roughly three quarters of the entire list, have an annual count of zero or unknown.

So there are two very different populations sharing one statistic. The 217 are the state as citizens actually experience it: tax, customs, social insurance, the things everyone touches. Those are 83.9% online, and 85.2% of the filings arrive that way. The rest is a long tail of procedures that exist in law and almost never fire.

When a government says half our procedures are digitised, half of what is doing all the work. Counting procedure types treats a customs declaration filed billions of times a year and a licence applied for twice as one each. No product team would report usage that way and survive the meeting.

The agency knows this, incidentally. It puts the 0.3% and the 95.6% on the same slide. I like that more than I expected to.

The reason given for the rest

For procedures that have not been digitised, ministries were asked why. Counted by procedure type, the most common answer, at 40.6%, is that the cost benefit of digitising is small or unknown.

Counted by annual volume, that answer drops to 21.3% and the largest category becomes digitisation already planned, at 48.8%.

That gap is the whole story and it is a rational one. Ministries digitise what has volume. Where volume is unknown, so is the case for spending money, so nothing happens, so nobody ever measures the volume. It is not laziness. It is a loop.

The way out of a loop like that is not a bigger budget. It is a cheaper default. If standing up a basic online form for a procedure that fires forty times a year costs almost nothing, the cost benefit question stops being interesting. Which is roughly what a shared platform is for, and Japan is building several.

About 40% of it is email

Here is my favourite line in the whole document. Of the procedures counted as digitised, about 60% run through a ministry system. About 40% are done by email.

Email is better than a fax and better than a train ride to a counter. I am not going to pretend otherwise. But nothing in an email is validated, nothing is a structured record, and the person receiving it retypes what you sent into whatever system the office actually runs on. The queue, the errors and the delay all still exist. They moved into a building you cannot see.

This is the government version of a pattern I see in companies constantly. A process gets called digital because the transport changed. The form is a PDF now. The PDF arrives by email. Someone opens it and types it into the system. Everyone reports the project as complete.

Where the online form actually breaks

Two findings in the survey point at the same failure, and both are design failures rather than technology ones.

The first is money. Among procedures that charge a fee, 61.9% of procedure types can only be paid offline. By volume it is better, just under half, because the high traffic ones were fixed first. Still: you can file online, and then you go to a bank counter or an ATM to pay. The report says out loud what that does. If the payment is offline, the convenience of the electronic filing drops, and paper filing keeps happening. It also notes that on the government's side, matching payments to applications by hand removes the benefit for staff too.

I have watched teams spend two quarters on a beautiful application flow and leave the payment step to whatever the finance system already did. It is the same mistake at national scale. A flow is only as complete as its last step, and the last step is almost always the one that takes money.

The second is paper originals. The survey looked at what attachments each procedure demands and how that correlates with being online. Procedures that require a family register, the 戸籍, are online 39.3% of the time. Procedures that require financial statements are online 77.4% of the time.

The difference is not importance or sensitivity. It is that one of those documents is habitually a certified piece of paper you fetch from a municipal office and the other is a file that already exists on somebody's computer. Require a paper original anywhere in a flow and you have made the whole flow paper, no matter how good the screens are.

Japan's answer to this is the registry work now underway, so that agencies can fetch corporate and property registry data themselves instead of asking the applicant to carry it. That is the correct fix. It is unglamorous, it takes years, and it will do more for the experience than any redesign.

The number I would put in front of a board

National procedures, citizen or business to the state: about 72% available online, about 80% of filings arriving that way.

Procedures where the receiving side is local government: about 40% available online, and about 30% of filings arriving that way.

The agency flags the caveat itself, that local figures cover only what ministries can see and exclude each municipality's own procedures. Take it as directional. The direction is still a forty to fifty point gap between the two governments a resident deals with, and the local one is the government you meet when you move house, register a child, or need a certificate. It is the tier with the least budget, the least specialist staff, and the most face to face contact.

Anyone selling into Japanese public sector work should read that pair of numbers as the map. The national tier is largely done and is now optimising. The local tier is where the actual experience of being governed here still lives, and it has barely started.

What I take from it as a designer

Three things I would carry into any digitisation project, public or private.

Count by volume, never by feature. Half our forms are online is a sentence that survives only until someone asks which half.

Follow the flow past the point where your team's responsibility ends. Payment, attachments, and the person retyping the email are all inside the user's experience even when they sit outside your org chart.

Be suspicious of any process described as digital where you cannot name the system that receives the data. If the answer is an inbox, the work is not done.

The Digital Agency deserves credit for publishing a document that makes its own progress look complicated. Most organisations would have led with the twenty point improvement and left the 217 out.

Sources

• Digital Agency, 行政手続等の調査結果概要, 22 July 2025. Published under article 25 of the act on promoting the use of information and communications technology in administration. Scope: every procedure defined in legislation administered by 26 national bodies. Phase 1 as of 1 October 2024, phase 2 as of 31 March 2024. Procedure types 75,085 against 61,390 in the FY2021 survey, and about 4.3 billion transactions a year.
• Same document: page 4 for the volume distribution and the 217, page 5 for the split between ministry systems and email, page 6 for the reasons given where procedures are not online, page 7 for fee payment, pages 8 and 9 for required attachments, page 10 for the national and local split.