AI Made 91.6% of Japanese Companies Faster and 3.9% of Them Richer
IPA, Japan's Information technology Promotion Agency, runs an annual survey on how companies here are doing with digital transformation. The 2026 edition landed in July, built on 1,799 responses collected between April and June.
Two numbers from it have been sitting in my head since.
Asked what concrete effect AI adoption had produced, 91.6% of companies said work became more efficient or faster. 3.9% said revenue or profit went up.
The rest of that list
It is worth seeing the whole shape rather than the two ends.
Work is more efficient or faster: 91.6%. Quality and speed of proposals improved: 48.9%. Overtime went down: 29.2%.
Customer satisfaction improved: 4.5%. The customer base expanded: 2.7%. Revenue or profit improved: 3.9%.
There is a wall in the middle of that list and it is exactly where the customer starts. Everything on the near side of the wall is large. Everything on the far side is a rounding error.
The usual explanation is that these things take time and revenue lags. I do not fully buy it, because the survey also asks what people are using AI for, and the answer explains the gap without needing a lag.
Look at what it is pointed at
The top three uses: summarising, translating and proofreading documents and audio at 82.5%. Writing documents and reports, internal and external, at 80.5%. Searching, collecting, analysing and reporting information at 77.0%.
Now the bottom: improving the company's own products and services, 10.9%. Supporting production, logistics and service delivery planning, 6.0%.
So roughly four in five companies point AI at their own paperwork, and one in ten point it at the thing the customer buys.
Given that, 3.9% revenue impact is not a disappointing result. It is the correct result. Nothing was aimed at revenue. The tool was deployed where the friction was most visible to the people choosing the tool, which is their own inbox.
This is the oldest pattern in enterprise software and AI has not changed it. The internal user is present at the decision. The customer is represented by a slide.
Which is a design problem, and I mean that literally
Consider what it takes to use AI on internal documents. You buy licences. People paste things in. Nobody has to agree on anything. There is no design work, no product decision, and no risk to the brand.
Now consider using it in the product. Someone has to decide what it does, where it appears, what it says when it is wrong, what happens to the user's data, how much latency is acceptable, whether the output is presented as fact or suggestion, and who is accountable when it misleads a customer.
Every one of those is an interface decision. Every one requires a person who can hold the customer's point of view in a room full of engineers and lawyers. If your company has no such person, the AI project will drift toward the version that needs no such person, which is summarising documents.
The survey has a number for this too, in a different section. 85.5% of companies report a shortage of people to drive digital transformation, in quantity, unchanged from 2024 and 2023. Three years flat. Whatever companies have been doing about that, it is not working.
The 31.8% that nobody quotes
Here is a number that cuts the other way, and I want to be fair to it.
Asked how AI performed against expectations, 31.8% said the effect met or exceeded what they expected. 50.6% said there was some effect.
So more than eight in ten got something. This is not a story about AI failing in Japan. The tools work. People use them. Overtime falls.
What has not happened is the second step, where efficiency turns into something a customer notices. IPA says this plainly across the whole report: results are high on efficiency items and low on items connected to creating company value, and this has not changed year over year.
That last clause is the alarming one. It is not a new gap. It is a stable one.
The size gap
AI adoption tracks headcount almost perfectly. Companies with over 1,001 employees are close to 80% adopted. Companies with 101 employees or fewer are at 16.6%.
A five to one spread. In a country where small and medium companies are most of the economy and most of the employment.
I have opinions about this that the data does not support, so I will keep them short. Small companies are where AI should produce the biggest relative gain, because they are the ones without a legal department, without a translator, without a research team, and without anyone whose job is writing documents. The tool substitutes for a role they never had the budget to hire.
What stops them is not licence cost. It is that nobody has time to figure out where it goes, and every vendor pitch is written for a company with a transformation office.
What I would actually do with this
If you are running one of the 86% who have AI somewhere in the business, here is the test I would apply.
Name one thing a customer of yours can see, touch or receive that is different because of AI. Not faster internally. Different, from where they sit.
If you cannot name one, you are in the 91.6% and not in the 3.9%, and no amount of additional internal adoption will move you across. The gap is not a volume problem. Doubling how much summarising you do will not produce a customer outcome, because summarising is not aimed at one.
Then pick a single customer facing surface and put the work there. Onboarding, support, search, the part of your product where people get stuck. Design it properly, with a designer, including what it does when the model is wrong, because that is the part that determines whether anyone trusts it twice.
One shipped customer facing thing teaches an organisation more than three years of internal rollout. It forces every conversation that internal use lets you avoid.
And it is the only way the 3.9% moves.
Sources
- Information technology Promotion Agency, 国内企業のDX動向・AI活用動向のポイントを公開, 16 July 2026. 1,799 responses from executives, information systems departments and DX teams at Japanese companies, collected 17 April to 12 June 2026. AI effects on page 5, uses on page 5, adoption by company size on page 4, effect against expectations on page 4, talent shortage on page 7.
- Information technology Promotion Agency, DX動向2026, the full report and data set.