AtifyWhen we started Atify, we worked mostly with large enterprises. It seemed the natural place to begin. They had budgets, internal technology teams, plenty of data and a real interest in what AI could do for their business, and we had experience implementing AI in complex corporate environments.
What we hadn’t expected was that more resources would not mean faster progress.
A single AI initiative at a large company can involve business leadership, IT, information security, legal, compliance, procurement and several operational teams. One department owns the budget, another the technology, another the process, and someone else makes the final decision. Ask who is actually responsible for making the implementation work, and the answer is often less clear than you would hope. Add the reviews a solution must pass before it reaches production, plus existing systems and internal policies, and it is easy to see why so many large companies start slowly. They write policies, evaluate platforms, run pilots and build internal capability, and many of those experiments never become part of everyday work.
That is not a permanent condition. Once a large company has clear ownership, approved technology and an implementation model it can repeat, it can scale very quickly. The preparation takes time, but the impact afterwards can be significant.
Smaller companies looked almost like the mirror image.
An owner or managing director can often approve a new idea in a single meeting. There are fewer layers, and changing a process is far easier. Willingness is rarely the problem. What small and medium-sized businesses (SMBs, called SMEs in OECD reports) tend to lack is the capacity and the expertise to make AI work in practice.
Many Japanese SMBs already use ChatGPT and similar tools to draft emails, translate, summarize documents, prepare marketing content or do research. But using AI now and then is very different from building it into a business. Asking AI to draft a reply to a customer is easy. Building a process that receives the inquiry, understands what the customer needs, retrieves the right company information, prepares a suitable response, brings in a person when necessary and records the result in the company’s system is much harder. It takes process design, technical integration, testing, data management and a clear view of where human judgment is still needed. This is where a lot of companies get stuck.
Yet the same traits that make this hard also make SMBs a very good fit for AI. Their processes are usually simpler, their decisions quicker, and the people deciding are close to daily operations. A well-chosen project can go from idea to production quickly, with results the business can see.
What the numbers say
The OECD surveyed more than 5,000 SMBs in seven countries, Japan included, in late 2024, and published the results in 2025 in Generative AI and the SME Workforce. Among the SMBs using generative AI, 65% said it improved employee performance, 35% said it helped them scale their activities, 29% said it helped them compete with larger companies and 26% reported higher revenue. About a third said it reduced the workload of staff or the owner, and 39% of those facing a skills gap said AI helped them compensate for it.
Fourteen percent had reduced their reliance on external contractors, but across all seven countries 83% reported no change in their overall staffing needs. For most early adopters, then, AI was strengthening the existing workforce rather than replacing it. These are self-reported results and they will vary by company and use case, but they suggest the value of AI for SMBs goes well beyond writing and summarizing.
Japan, meanwhile, is still early. In the same survey, 23.5% of Japanese SMBs were using generative AI, the lowest of the seven countries (Germany was highest at 38.7%). A separate survey commissioned by Rakuten and run by Edelman Data & Intelligence in late 2024 found that 16% of 300 SMB owners and decision-makers said their company used AI, a broader category than generative AI. Among those who did not, 40% could not see what AI could do for their business, 34% pointed to a lack of technical expertise, 31% worried about return on investment and 28% cited implementation costs. Those who did use AI mostly used it for copywriting (43%), routine process automation (30%), translation (23%) and software development (21%).
The two surveys used different samples and methods, so the figures should not be compared directly, and both are from late 2024, so adoption today may be higher. But the pattern is consistent: many Japanese SMBs have not yet built the expertise to identify and deploy the right AI solutions. That is an adoption gap, and it is also an opportunity.
Buying a tool is not the same as implementing it
There are now more AI products than anyone can keep track of, and nearly every business system claims to include AI. It is tempting to think implementation is easy: pick a product, buy a subscription, let people start using it.
In practice the technology is only part of the work. A company first needs a problem worth solving. It has to understand how the process runs today, where time is lost, what information is needed and what could go wrong, and then fit the solution into the systems and habits already in place. Skip that, and you end up with a handful of disconnected tools that employees try for a few weeks and quietly drop. A few individuals get a bit faster, and the business stays the same.
Even with the right problem, choosing the technology takes care. Rarely is there only one model or platform that can do the job. Two models can give similarly useful results at very different operating costs, and published API pricing shows how widely costs vary, even between models from the same provider. For a small pilot that hardly matters. Once a solution handles thousands of customer inquiries, documents or transactions, it can decide whether the project makes commercial sense.
The newest model is not automatically the best one. A simpler, cheaper model may be enough to classify inquiries or extract data from documents, while ambiguous requests or sensitive customer communication may call for something more capable. Then there is the question of whether to buy a product, customize a platform or build something new, and how each option fares on accuracy, response time, Japanese-language performance, data privacy, integration, vendor dependency and maintenance. The way to settle it is to test the options on the company’s real work and pick the simplest solution that meets the standard.
Mid-sized companies have an extra difficulty. They have more budget and complexity than a small business but not the specialist teams of a large enterprise, and a failed implementation is still costly. In our experience, what decides the outcome is a strong internal owner: someone who understands the business problem, coordinates decisions, works with the implementation team and makes sure the solution becomes part of daily operations. With that person, mid-sized companies can move remarkably fast. Without one, even a promising pilot loses momentum.
Why we are widening our focus
Working with large enterprises taught us that a successful transformation is about more than building the solution. It means connecting business needs, technology, operations and people, and keeping delivery disciplined: clear responsibility, managed dependencies, careful testing, measured results and continued improvement after launch. We learned this on large, complex programs. We now believe it can be just as valuable to Japanese SMBs.
They rarely need a large transformation program or an expensive custom platform. What they need is help with practical questions. Which problem should we solve first? Will AI really improve this process? Can we use an existing product, or do we need something tailored? Which technology gives the right balance of quality and cost? How should it connect to our current systems? What can AI safely access? Where should a person stay involved? And how will we know it is working?
We help companies answer these questions and then turn the answers into a working solution. We start with the business problem, not the technology, and we do not think every process needs AI. Sometimes plain automation is more reliable and cheaper to maintain, and often the best result combines automation, AI and human decision-making. A first project might cut the time spent on customer inquiries, help a sales team find and qualify leads, let employees search across company documents, automate repetitive administrative work or provide support outside business hours.
The first project does not have to transform the company. It has to solve a real problem and produce a result the business can see and measure. Once that works, the company can build from there.
Waiting is a decision too. Starting now does not mean a large, risky investment or overhauling the whole organization. It means building practical experience, finding valuable use cases, improving data and processes and developing the ability to use AI responsibly. Companies that begin now have time to learn from their mistakes. Those that wait for the technology to mature may find their competitors have spent years improving processes, cutting costs and building know-how that cannot be bought overnight. The tools will be available to everyone, so the advantage will come from knowing where to apply them and how to run them at scale.
A few years from now, AI capability will probably be a normal part of running a competitive business. Companies do not have to solve everything today, but they do need to start.
If you are thinking about AI in your business and are not sure where to begin, get in touch with us. You do not need a finished plan. Tell us about the process that slows you down, and we can take it from there.
References
- OECD (2025), Generative AI and the SME Workforce: New Survey Evidence, published 5 November 2025. Survey of more than 5,000 SMBs in Austria, Canada, Germany, Ireland, Japan, Korea and the United Kingdom, conducted in late 2024. https://www.oecd.org/en/publications/generative-ai-and-the-sme-workforce_2d08b99d-en.html
- OECD (2025), Artificial Intelligence and the Labour Market in Japan. https://www.oecd.org/en/publications/artificial-intelligence-and-the-labour-market-in-japan_b825563e-en.html
- Rakuten Group, “Rakuten Survey Reveals AI Awareness Gap and Growth Potential for Japanese SMEs,” 29 January 2025. Commissioned by Rakuten, conducted by Edelman Data & Intelligence among 300 SMB owners and decision-makers, 15 to 26 November 2024. https://global.rakuten.com/corp/news/press/2025/0129_01.html
- OpenAI, API pricing. Prices change over time and should be checked when evaluating a production implementation. https://openai.com/api/pricing/