Japan’s SMEs Want AI, but Practical Know-How Is Missing

The most revealing number in a new survey of Japanese small and midsize businesses may not be an adoption rate. It is 47.9%: the share of non-adopters who said they did not know which work AI could perform. The tools are visible. The missing layer is translation—turning a general-purpose model into a controlled change in quoting, purchasing, customer service, inspection or paperwork.

Japan’s adoption gap is also a comprehension gap

Of all 1,653 respondents, 22.7% said they understood AI’s mechanisms and benefits deeply enough to know how it could fit their work. Another 43.8% knew the outline but not how to apply it to their own business. The remaining respondents said AI was irrelevant to them or that they did not know.

The institute then asked the 1,099 companies in the first two groups what they were doing. It found 36.9% already using AI, 20.1% considering or testing it, 34.2% unable to introduce it despite seeing a need, and 8.7% with no intention to adopt. The correct conclusion is not that 36.9% of all Japanese SMEs use AI. It is that, even inside a relatively aware group, knowledge of application sharply separates users from stalled companies.

The quiet office tasks came first

The first wave is not a factory without workers. Among 303 respondents who said they both understood AI deeply and already used it, 71.9% selected writing, summarizing and proofreading as a use case. Information gathering and research followed at 61.7%, idea generation at 52.8%, and sales documents and proposals at 50.5%. Respondents could choose multiple answers.

The reasons were equally practical: 75.9% cited efficiency and reduced work hours; 67.7% cited productivity. New revenue and competitive differentiation ranked much lower. Japan’s Information-technology Promotion Agency reached a compatible conclusion in its broader DX Trends 2026 study: AI is spreading, but its measured effects remain concentrated in speed and efficiency rather than new value or enterprise transformation.

Three questions before the subscription

These figures complicate the familiar claim that small companies are held back mainly by cost. Money matters, but it did not top this survey. A cheap model cannot decide which process deserves attention, what information may leave the company, how errors will be caught or whether a saved hour is worth the new review burden.

The useful first purchase is often not an AI license. It is a baseline: how long one task takes today, how often it is wrong and who is responsible for checking the result.

Success rates drawn from a selected group of users

The report says 91.7% of the 303 respondents who deeply understood and already used AI perceived an improvement in operating efficiency, while 86.5% perceived fewer work hours. Those are encouraging self-reports from a selected group. They do not measure every attempted deployment, and they are not an audited estimate of economy-wide productivity.

A defensible pilot records the old process before the tool arrives. It measures minutes per document, rework, error frequency, response time or orders completed. It also counts the time required to verify generated material. Without that comparison, enthusiasm can be mistaken for a return.

The employee account outside the company’s view

Governance is already chasing behavior. Within that same 303-company group, 47.2% said they used a company-managed AI service, 38.0% used a service registered personally, and 13.5% used both. Personal accounts make experimentation easy, but they can leave managers unable to see what customer information, designs, prices or internal documents have been submitted.

A small company does not necessarily need a 100-page policy. It does need specific boundaries: prohibited data, approved services, required human checks, retention rules and an owner for incidents. A blanket ban can push activity out of sight. A bounded experimental area gives employees somewhere safer to learn.

From office computerization to generative AI

Japanese small businesses have crossed several digital thresholds before. Accounting and payroll were computerized; personal computers and spreadsheets moved calculations onto desks; the internet changed customer communication; packaged software standardized workflows; cloud services lowered the cost of servers and maintenance. Each shift promised efficiency, but each also required work to be described in a form the system could process.

In 2018, the Ministry of Economy, Trade and Industry’s DX Report warned that aging core systems could prevent corporate transformation, popularizing the phrase “the 2025 cliff.” The policy conversation then expanded from replacing systems to using data and technology to change operations and business models. The 2026 Small and Medium Enterprise White Paper now discusses AI transformation, or AX, as an opportunity for growth.

Generative AI is unusual because employees can begin before management completes a technology project. A browser and a personal account may be enough. That reduces the cost of experimentation while increasing the importance of governance, training and verification.

What a small company can do on Monday

Of 1,003 respondents asked about future intent after excluding companies with no plan to adopt, 29.8% expected to push AI substantially and 53.4% somewhat. The combined 83.2% measures stated intention, not a budget or deployment.

Japan’s labor shortage gives small firms a real reason to reduce repetitive work. But the survey suggests the decisive infrastructure is not only computing power. It is the ability to select a task, protect information, test a result, teach employees and stop a bad experiment. AI may be general-purpose technology. Implementation remains stubbornly specific.