Skip to main content
2026.09.01How labour administration works

Does "one person, 60 client companies" work in labour too?

Joji Uramatsu

Joji Uramatsu

Shakai Hoken Roumushi (Certified Social Insurance and Labor Consultant), Gyoseishoshi (Certified Administrative Procedures Legal Specialist), Registered Real Estate Transaction Specialist — 四葉社会保険労務士事務所/四葉行政書士事務所

A tax accountant running 60 client companies with zero staff has become a talking point. But the first thing he did was not automation — it was drawing the line between what gets automated and what a human looks at. The same is possible in labour, but the line falls in a different place. Accounting data flows in on its own; in labour, a fact only comes into being once someone tells you about it.

In short: a tax accountant who runs 60 client companies with no staff has become a talking point. But the first thing he did was not automation — it was drawing the line between what gets automated and what a human looks at. The same can be done in labour, but the line falls in a different place.

"Will AI do away with the work of licensed professionals?" We have been asked this many times over the past six months. Answering in the abstract gets us nowhere, so let us start from what is actually happening.

What is actually happening?

畠山謙人税理士事務所 — the Kyoto City office of the tax accountant 畠山謙人 (Kento Hatakeyama) — was featured in a customer case study published by the accounting software company freee. The office consists of one representative tax accountant and zero staff. On that footing it has more than 60 retained client companies, with enquiries still increasing. Its clients are mainly seed-stage startups, and the working day ends at 17:00. Its posts on social media have recorded 3.5 million impressions.

What supports this is a data integration between Claude Code and freee. The AI retrieves the unposted transaction lines that have accumulated in the accounting software and works through the journal entries. The tax accountant, the case study says, has moved from being "the person who does the work" to "the person who reviews it".

Looked at as figures alone, it may sound like a distant world. That is not what this article is about.

What was the first thing he did?

It was not automation. It was drawing a line.

According to the case study, the first thing 畠山 worked on was creating a configuration file setting out what the AI may do and what it must not do. He himself calls it "the office's constitution."

On that footing, he says the following (the original is in Japanese; the English is our translation).

What is most frightening about automated journal entries is "posting an entry for something that should never have been posted at all"

What comes first is not processing quickly, but not letting the machine touch what must not be touched. That, as we read it, is the most important part of this case.

He then goes on, on the division of roles between AI and people (our translation).

The checking of AI output and the final judgement should always be done by a human. […] If the tax accountant does not carry that responsibility, there is no point in holding the licence.

What is different between accounting and labour?

This is the heart of it. It does not follow that "if a tax accountant can take 60, a shakai hoken roumushi can take 60 too." The structure is different.

Accounting (bookkeeping)Labour and social insurance procedures
Where the input comes fromBank and card transaction lines. Machine-readable data flows in automaticallyA hire, a departure, a change in dependants, sickness or injury, childbirth, a change of address. It arises only once a person tells you
CycleThe same processing every month. Easy to reduce to a patternDriven by events. You do not know when they will happen. Annual filings (算定基礎届, the standard remuneration base notification) and as-and-when filings (月額変更届, the monthly-remuneration change notification) sit side by side
What the judgement is aboutWhich account to post to. A pattern can be learned from past entriesEstablishing the facts. Is this person a worker? Does a relative living in the same household meet the requirements to become an insured person? You cannot know without asking a person
How errors surfaceThere are chances to notice at the monthly close or at the year-endIt surfaces years later, in the person's own benefits — a pension, or unemployment benefit. And the correction has to be made retroactively
How the law changesAmendments are frequent, and the effective dates are staggered. The company-size requirement for social insurance, for example, comes down in four stages — October 2027, October 2029, October 2032 and October 2035

The biggest difference is the first row.

Automating accounting was a question of "how to classify the data that has flowed in". Bank transaction lines arrive every day whether or not you do anything.

Labour is different. A new dependant, a change of address — you do not know about either unless someone tells you. If the employee forgets to mention it, everything stops there. However clever the AI, it cannot process data that never flows in.

The fourth row bites as well. A wrong journal entry can be caught at the year-end close, but a missed filing may not come to light until the person reaches the age of drawing a pension. By then, years have passed.

Where in labour would you hand work to AI?

So what should be handed to AI in labour is not the equivalent of journal entries.

It is managing deadlines, and making sure nothing goes unasked.

  • Once a hire is scheduled, laying out what has to be filed and by when
  • At 算定基礎届 season, picking up every person in scope without omission
  • Re-sorting the effective dates of amendments in the order in which they touch your own company
  • Turning "the things people forget to mention" into questions we go out and ask

None of these is a judgement. They are the work of preventing gaps, upstream of judgement. This is where AI is strong — and it is also where most accidents happen in labour.

Conversely, what must not be handed over is establishing the facts, and the judgement that follows from them. Neither the line between outsourcing and employment nor whether a relative living in the same household can join employment insurance is settled without hearing out how things actually work. Ask an AI and a plausible answer comes back. That is exactly why it is dangerous.

Why must a human make the final judgement?

What 畠山 calls "the point of holding the licence" is, for shakai hoken roumushi, written straight into the statute.

The Shakai Hoken Roumushi Act (社会保険労務士法, Act No. 89 of 1968), Article 2, paragraph 1, item 1 defines the "applications and other documents" that make up the reserved work as follows (our translation).

applications, notifications, reports … and other documents (including, where an electromagnetic recordis prepared in place of preparing such a document, that electromagnetic record)

It says "including … that electromagnetic record." In other words, even where an AI produces the data for an electronic filing, that data is itself the "application or other document" the statute speaks of. Paper or data, the treatment does not change.

On that footing, Article 27 draws the line (our translation).

A person who is not a shakai hoken roumushi or a shakai hoken roumushi corporation must not, in response to a request from another person and for remuneration, carry out as a business the work listed in Article 2, paragraph 1, items 1 through 2

What decides it is not who prepared the document, but whether it is done for another person, for remuneration, as a business. Doing your own company's filings yourselves needs no qualification. That holds whether you use software or AI. We set this out in detail in We've brought in freee HR. So what does a retained shakai hoken roumushi actually do?.

The cleverer AI becomes, the more the meaning of a qualification shifts from "being able to produce it" to "taking it on." That, as we read them, is what 畠山's words mean.

What can 四葉社会保険労務士事務所 do?

This office draws the line on the same principle.

We use AI for organising materials, surfacing the issues, checking wording against the statute, and sorting deadlines. Judgement is made by the qualified professional. We have also decided in advance how far client personal data may be passed to outside services. That practice is set out concretely in Can I ask an AI about a labour question first, and then consult a shakai hoken roumushi?.

And here is what we most want to say in this article.

What 畠山 did was reduce the work inside his own office. What this office puts its effort into is reducing the work inside the client's company. We support the build-up to processing payroll in-house on freee, included in the retainer fee. There is no separate charge. For detail, see We want to run payroll in-house on freee — what do we ask a shakai hoken roumushi for?.

Taking the work on ourselves would earn this office more. We still put in-house support inside the retainer because we have defined the retainer as the price of consultation. Less work does not mean less need to consult. If anything it increases — as consultation on the design of the settings, and on each month's close. The reason for this design is in What is a shakai hoken roumushi's retainer actually paying for?.

Consultation is free of charge, the first time and every time after. The vague stage — "could we do something with AI too?" — is perfectly fine.

Frequently asked questions

Q. If you use AI, does the retainer get cheaper?
A. Because this office sets the retainer as the price of consultation, the retainer does not change when the amount of work goes down. On the other hand, fees that are set by volume of work — payroll, for instance — do go down if you switch to in-house. Stop the outsourcing (1,100 yen per employee per month, tax included) and process it yourselves, and that part disappears. The support for switching is itself included in the retainer, so we charge nothing extra for the support.

Q. Won't a shakai hoken roumushi's work be replaced by AI in the end?
A. We expect the volume of work to fall. But labour has establishing the facts at the centre of it, and that does not move forward without asking people. The law also provides that carrying out procedures for another person for remuneration is not open to anyone who is not a shakai hoken roumushi (社会保険労務士法, Article 27). Rather than being replaced, what changes is where the qualified professional spends time — that is how we see it.

Q. Is it a nuisance if I bring in an answer an AI gave me?
A. Not at all. In fact, coming in after you have looked it up makes the consultation faster. But AI answers can drift on article numbers and effective dates. Bring them as something to be checked against the source.

Q. Can a company as small as ours really go in-house on freee?
A. The fewer employees you have, the more easily the benefit of going in-house shows: with few people, the monthly processing itself is not heavy. What trips companies up is not the operation but the design of the initial settings. Build that part with us, and most companies can run the rest themselves. Start by sorting out whether outsourcing or in-house suits you better — please do get in touch.

Sources for this article

  • freee, 「スタッフ0人、顧問先60社、17時退勤。一人税理士が実践する『責任あるAI』」 (customer case study, 畠山謙人税理士事務所). Viewed 14 August 2026. The size of the office, the number of client companies, the methods and the remarks quoted all come from this published case study
  • The quoted remarks are in Japanese in the original; the English here is our translation, with nothing added and nothing removed. We have written no speculation about his intentions beyond what is published
  • 社会保険労務士法 (Shakai Hoken Roumushi Act, Act No. 89 of 1968), Article 2, paragraph 1, item 1 and Article 27confirmed on e-Gov法令検索 on 14 August 2026. The English of the statutory text quoted above is our translation
  • The staged reduction of the company-size requirement for social insurance follows 社会経済の変化を踏まえた年金制度の機能強化のための国民年金法等の一部を改正する等の法律 (Act No. 74 of 2025) and materials published by the Ministry of Health, Labour and Welfare
  • The comparison between the nature of accounting work and that of labour and social insurance procedures is our own analysis. It is not laid down as such by the system
  • 畠山謙人税理士事務所 is a tax accountant's office and has no relationship whatsoever with this office. We have taken up a published case study entirely on our own initiative

This article does not go so far as to decide whom you should consult. Bookkeeping, closing the books and tax returns are the work of a tax accountant. This office does not handle them, so we will point you to a tax accountant and you engage them directly. This office does not accept referral fees. Labour and social insurance procedures, work rules, payroll and support for bringing it in-house are handled by 四葉社会保険労務士事務所. Fees are set out in the fee schedule, and the questions we are asked most often are collected on the FAQ page.

This article is general information. Judgments that fit your particular circumstances are made by a qualified professional after a meeting. Written by Joji Uramatsu (Shakai Hoken Roumushi (Certified Social Insurance and Labor Consultant), Gyoseishoshi (Certified Administrative Procedures Legal Specialist), Registered Real Estate Transaction Specialist).

Let’s start by sorting out where things stand.

四葉社会保険労務士事務所 (Kohinata, Bunkyo-ku; a 5-minute walk from Myogadani Station on the Tokyo Metro Marunouchi Line) helps you, starting with a review of your current labour practices.

LINE connects you directly to our representative, Joji Uramatsu. Messages are accepted 24/7 and answered in order.

5 min walk from Myogadani Sta. (Tokyo Metro Marunouchi Line)|Tue & Wed 10:00–19:00 / Mon, Thu–Sun 18:00–19:00