67.1% of Japanese Companies Trigger AI Manually | 3 Reasons Automation Isn't Moving Forward

What you'll learn in this article:

• The reality that 67.1% of Japanese companies trigger AI manually
• Why automation isn't advancing despite high AI adoption rates
• The serious challenge of 50.7% of companies having unorganized internal data
• The global rise of "AI agentification" and Japan's lag behind
• The path toward "autonomous AI operations" that companies should pursue

Two-Thirds of Japanese Companies Trigger AI "Manually"

A surprising reality came to light in the "Japan AI Operations Report 2026 Summer," released in June 2026 by Mer Inc. It revealed that 67.1% of Japanese companies using generative AI still trigger it through "manual human activation."

The survey was conducted across 548 companies with 100 or more employees. The results showed that AI activation methods broke down as follows:

• Humans manually running a set prompt: 36.3%
• Entering instructions via chat: 30.8%
• Automatic scheduled activation: 16.6%
• Triggered by data changes: 12.0%

In other words, at 67.1% of companies, someone has to "ask" AI to work every single time. This means that AI, which is supposed to "assist with human work," is instead being "looked after by humans."

Adoption Rate Is 87%—So Why Isn't Automation Advancing?

What makes this particularly striking is that generative AI utilization among Japanese companies had reached 87% as of spring 2026. That is, while nearly all companies have "introduced" AI, they haven't progressed to the stage of "running it automatically."

The primary cause is the lack of organized internal data. The survey found that 50.7% of companies responded that "internal data such as customer information and project information is not organized."

Without organized data, AI cannot make decisions autonomously. For example, if you try to build a flow where "a proposal is automatically generated when a new project is registered," AI won't know where to look if project data is scattered across different locations.

Even more serious is the tracking of AI usage. Only 13.0% of companies can track AI usage across the organization—meaning nearly 90% of companies cannot determine "who used which AI, when, and to do what."

Even After AI Output, 67.4% Still Rely on Manual Work

The problem doesn't stop at activation. When it comes to handling AI-generated output, manual work makes up the majority here as well.

• Manually processing output into deliverables: 46.2%
• Manually copying and transcribing: 21.2%
• Automatically applied after approval: 23.7%
• Automatically processed based on conditions: 3.5%

A combined 67.4% are "copy-pasting" or "manually revising" AI output. At this rate, companies that think they're saving time with AI may actually just be creating more work for people.

Tomoaki Sawaguchi, Representative Director of Mer, points out that "what has spread is AI's 'applications,' not the 'mechanisms' by which AI operates." In other words, while "ways of using" AI—such as writing text with ChatGPT or creating designs with image-generation AI—have become widespread, the infrastructure to integrate AI into business workflows has not been established.

"AI Agentification" Is Advancing Rapidly Worldwide

Meanwhile, globally, the shift toward "AI agents (AI that operates autonomously)" is accelerating. It is predicted that by 2026, AI agents will be embedded in 80% of enterprise applications, with market growth rates expected to exceed 46% annually.

An AI agent is AI that makes its own decisions and acts without human instruction. For example, when a customer sends an email, it automatically analyzes the content, references past interactions, drafts a reply, and requests confirmation from the person in charge—all as part of a seamless automated flow.

Global companies have begun adopting "multi-agent systems," in which AI agents coordinate complex business workflows spanning multiple departments such as sales, support, supply chain, and accounting.

While 67.1% of Japanese companies are still triggering AI manually, the world is moving toward "AI operations with minimal human intervention."

Still, 67.2% of Companies Plan to Expand AI Investment

Despite the challenging current state, there is reason for optimism. The survey found that 67.2% of companies responded that they plan to "expand AI investment going forward."

The top three investment priorities cited were:

• Employee training and education: 48.9%
• Guideline development: 48.7%
• Business workflow redesign: 18.4%

Notably, "training" and "guidelines" top the list. This indicates that many companies recognize the challenge not as "AI's own capabilities" but as "insufficient preparation on the user's side."

In fact, data shows that companies where executives themselves use AI implement it 2–3 times faster. This means that rather than leaving it to frontline employees, executive-level commitment is what determines whether AI adoption succeeds or fails.

The Key to Solutions: An Operations Model That Doesn't Require "Using AI"

So how can Japanese companies break free from "manual activation"?

Mer's report proposes a shift to an "autonomous operations model." Rather than employees "using AI," the idea is to build "a mechanism where AI operates on its own."

Concretely, this involves the following steps:

• AI automatically activates triggered by business events (project registration, inquiry received, etc.)
• AI automatically retrieves necessary data and executes processing
• Results are automatically reflected in relevant systems (approval flows remain in place)
• People focus exclusively on situations requiring judgment and on handling exceptions

Under this model, employees do not need to give AI instructions. AI operates naturally within the flow of work, and people can concentrate on tasks that only humans can do—such as "approving AI proposals" or "making judgments on unusual cases."

However, this requires certain preconditions: organized internal data and system integration. As long as the "unorganized data" problem affecting 50.7% of companies remains unsolved, autonomous operations are impossible.

3 Steps to Take Right Now

So where should companies that want to advance AI automation start? Drawing on the report and expert recommendations, here are three steps.

Step 1: Fully automate one business process

The reason many companies fail is that they try to change everything at once. Start by selecting a single business process with "clear inputs, clear outputs, and defined human checkpoints," and focus exclusively on automating that one process thoroughly.

The key to success is starting simple—for example, "initial responses to inquiry emails" or "automatically generating weekly reports from daily reports."

Step 2: Consolidate data in one place

For AI to operate automatically, data must be organized. Begin by identifying the data used in the workflow you want to automate, and consolidate it into a single database or system.

If Excel files and Google Sheets are scattered everywhere, AI cannot make any judgments. Establishing a clear "address" for your data is essential.

Step 3: Design logs and approval flows

Once AI begins operating automatically, it becomes harder to see "who did what." That's where AI usage logging and approval flows for important decisions become necessary.

The survey found that only 11.3% of companies invest in log and approval flow design. Without this, however, it becomes impossible to prevent AI from running out of control or to investigate the cause of problems when they arise. This is something that must be established in tandem with automation.

Conclusion: From Being "Used by AI" to "Making the Most of AI"

What this survey brought into sharp relief is the reality that while many Japanese companies are using AI as a "tool," they have not embedded it as a "mechanism."

• 67.1% of companies activate AI manually
• 50.7% of companies have unorganized internal data
• Only 13.0% can track AI usage across the organization
• AI agentification is advancing rapidly worldwide
• The key to solutions is the shift to an "autonomous operations model"

There is a significant gap between introducing AI and actually producing results with AI. Yet, as reflected in the 67.2% of companies planning to expand investment, many companies are beginning to recognize this challenge.

What companies need going forward is not the perspective of "how to use AI," but "how to build mechanisms by which AI operates." Rather than people looking after AI, AI supports people's work. The time to build that foundation—toward that kind of future—is now.

---

This article is a cross-post from AI Friends.