You Bought the AI Tool and Nobody Uses It

When a company asks me about generative AI, the situation is almost always identical. The contract is signed. Accounts have been distributed. Some people are using it. There is nothing to put in front of the board.

If you file that under "Japan is behind on AI", you will buy the wrong solution. Read the numbers in order and the thing Japan is behind on is not adoption.

The usage figures are not what people quote

The Ministry of Internal Affairs and Communications publishes the Information and Communications White Paper. In the 2024 survey, 26.7% of people in Japan had used a generative AI service. The United States was 68.8%, Germany 59.2%, France 81.2%.

That is the paragraph everyone reproduces. Now add the year before. Japan was at 9.1% in 2023, so it roughly tripled in twelve months. The US came from 46.3%, Germany from 34.6%. In relative terms Japan moved fastest of the group. It started from further back.

The company level data is stranger still. In IPA's DX Trends 2025, among firms with more than 1,000 employees, the share that have actually deployed generative AI is highest in Japan of the three countries surveyed. Add trial use and Japan is still highest. At large company scale, Japan is not losing on adoption.

The gap sits on one survey line

The same study asks how generative AI is actually being used.

For "individuals or departments trialling it" and "individuals using it in their own work", all three countries answer at similar rates. One option separates them: whether the tool is built into the department's work process. That is where Japan's response rate drops.

That is the whole story. The tool arrived. Individuals use it. Nothing about the work changed.

On the ground it looks exactly like that. People use it for first drafts, translation and summaries. Genuinely useful. But no step has been removed from any process, so nothing appears in any number. When leadership asks what the return has been, the person responsible is not failing. There is simply nothing measurable to report.

Why it never reaches the process

Three reasons, and none of them are technical.

Embedding a tool in a process means deciding whose job changes. This check is no longer required. This approval moves. This role does something different from next month. An IT department cannot make that call. IPA describes Japan's pattern as inward facing and partially optimised, with notably weaker links between management, IT and business units than in the US or Germany. When three departments sit between the person who could remove a step and the person using the tool, the step stays.

The second reason is aim. IPA found Japanese firms understand DX results in terms of cost reduction and efficiency, where American and German firms report them in revenue and profit. Point AI at efficiency and today's process gets slightly faster. That is worth something, and its ceiling is today's process.

The third is that nobody owns it. More than 80% of Japanese companies report a shortage of DX people. And 19.4% are carrying out no talent acquisition at all. Some of the companies reporting the problem are doing nothing about the problem, which means the constraint is not the labour market.

What has actually worked

The projects that worked did not start from the tool.

Pick one process that happens every week and that somebody complains about. It does not need to be strategic. One nearby annoyance is enough.

Write down what happens now, step by step, including who is waiting for whom. Almost no team has done this, and the resulting page usually looks bad. That is the useful part. You will reliably find two or three steps that can only be explained by a system limitation nobody has revisited in five years.

Then ask which step disappears, not which step gets faster. Faster is a rounding error. Disappearing changes the shape of the work, and only the second one shows up in a result.

Change one step. This is the rule teams break most often. The pilot becomes a redesign of a whole department, takes nine months, and dies inside an approval process before reaching a customer.

Then name one person. Not a project team. A person who has both the authority to change the step and the obligation to report what happened. A surprising number of companies have handed over one of those two and not the other.

Smaller companies are a different problem

The gap by company size may matter more than the gap between countries.

Among Japanese firms with 100 or fewer employees, "interested but no plans" and "no plans at all" together account for close to 80% on generative AI. In the US that group is just over 20%, in Germany just under 40%. Across DX generally, 96.1% of Japanese companies above 1,000 employees are doing something, against 46.8% of those under 100.

But the reasons given do not read like resistance. The most common answer is that they do not understand what benefit DX would bring to their company. The second is that they lack the knowledge and information to judge.

That is an unanswered question. It has gone unanswered because most of the material aimed at these companies explains what the technology is rather than what changes in their office on a Tuesday morning.

Nobody needs to understand the technology to make this decision. They need one line describing which step in their own work disappears.

The short version

Japan does not have a generative AI adoption problem. Individual use tripled in a year, and its large companies deploy at the highest rate of the three countries compared.

It has a work design problem, which existed before generative AI and will outlast the current generation of models. What AI did was hand every company the same capable tool and then show publicly which of them could change anything about how they work in order to use it.

References

• Ministry of Internal Affairs and Communications, 令和7年版 情報通信白書, individual AI use, July 2025
• Ministry of Internal Affairs and Communications, 令和7年版 情報通信白書, enterprise AI use
• IPA, DX動向2025 and the full report PDF, figures 1-2, 1-4, 2-10, 2-11 and 3-5