Conclusion

To put it simply, most small and medium enterprises do not need a ¥3 million AI agent.

AI agents are all the rage. “They autonomously make decisions and manage operations”—with such promises, a new wave of AI agent services priced at around ¥3 million per month is emerging.

But I want to ask: Does your company really have ¥3 million worth of “decision-making”?

From my experience observing local small and medium enterprises, about 80% of operations consist of “tasks” rather than “decisions.” Transcribing orders, checking inventory, reconciling invoices, and aggregating daily reports—these are not tasks that require advanced AI agents. They can be handled adequately with ¥50,000 worth of script automation.

In this article, I will compare the realities of ¥50,000 automation and ¥3 million AI agents with concrete numbers. I will clarify the “granularity of systematization” that small and medium enterprises should truly choose.

—

Patterns of “Glamorous Failure” for ¥3 Million AI Agents

First, let’s look at a case of AI agent failure.

A mid-sized manufacturing company (80 employees) introduced an AI agent with the aim of automating its sales process. The monthly cost was ¥2.8 million. It could handle many tasks, including generating estimates, prioritizing customer responses, and automatically sending follow-up emails.

What was the outcome?

Usage dropped to 15% within three months.

The reason is simple. Sales representatives felt they could not trust the priorities set by the AI and ended up relying on their instincts. The automatic estimate generation could not reflect subtle nuances (like discount rules for specific clients or past verbal agreements), requiring manual adjustments each time.

¥2.8 million x 3 months = ¥8.4 million. The results achieved were almost zero.

This is not an uncommon story. Failures in AI agent implementation generally share a similar structure:

1. They tried to delegate “decision-making”—but the tacit knowledge required for those decisions had not been verbalized.
2. The goals were vague—they started with the abstract objective of “improving operational efficiency.”
3. The field did not use it—the decision to implement was made by management, but the users were on the ground. This gap was never bridged.

In short, AI agents are introduced to replace “personalized decision-making,” but they cannot function if that decision-making is not verbalized and structured in the first place.

There are things to do before paying ¥3 million.

—

Reasons Why ¥50,000 Automation Continues to “Quietly Win”

Now, let’s introduce a contrasting case.

A local food wholesaler (12 employees) had a staff member logging into three different client systems every morning to transcribe the previous day’s order data into Excel. This took about 90 minutes each day, totaling around 30 hours per month.

This was automated using a Python script and Google Apps Script. The initial development cost was ¥150,000, with a monthly operational cost (cloud usage fees + maintenance) of about ¥30,000.

Result: 30 hours of work per month became zero.

If we assume the employee’s hourly wage is ¥2,000, that’s a reduction of ¥60,000 in labor costs per month. Even after deducting the ¥30,000 operational cost, that’s a pure cost saving of ¥30,000. While this amount may seem small at first glance, the true value lies elsewhere.

The employee began to “analyze order status first thing in the morning and proactively arrange for products at risk of stockouts.” The 90 minutes previously spent on transcription transformed into time for decision-making and action. As a result, opportunity losses due to stockouts decreased by 35% compared to the previous year.

Another example:

A trading company dealing in construction materials (25 employees) implemented a script incorporating an AI forecasting model for inventory management. This system automatically generates monthly demand forecasts based on three years of shipping data. The monthly cost is about ¥50,000 (for cloud ML infrastructure + data integration script maintenance).

Inventory turnover improved by 1.4 times, freeing up approximately ¥4 million in cash tied up in excess inventory annually.

Investing ¥50,000 x 12 months = ¥600,000 annually yielded a return of ¥4 million. The ROI is about 6.7 times.

What these cases have in common is that they automated “tasks” rather than “decisions.” And the humans freed from tasks improved the quality of their decision-making.

—

The Correct Approach to Systematization is Determined by “Granularity”

At this point, I want to clarify. The difference between ¥50,000 and ¥3 million is not just a cost difference. It’s the difference in the granularity of automation.

For small and medium enterprises, the first step should be “small-granularity automation.”

There are three reasons for this:

1. The cost of failure is overwhelmingly small. If ¥50,000 automation fails, the pain is limited. A failure costing ¥3 million could be fatal for a small or medium enterprise.

2. Success experiences accumulate in the field. Success in small automation prompts the field to think, “Where else can we automate?” This raises the organization’s AI literacy significantly. It is far more effective than top-down training.

3. The structure of personalization becomes visible. When trying to automate tasks, the question “Why is this task done in this way?” inevitably arises. It is only then that the structure of previously personalized operations becomes visible. Systematization is the reverse of personalization. Implementing an expensive AI agent without understanding the nature of personalization is meaningless.

—

“Transparency” Becomes a Weapon for Small and Medium Enterprises

Another often-overlooked point is the transparency of AI workflows.

When large companies introduce AI, it can cost several million yen just to establish a governance structure. Audit compliance, accountability, bias checks—due to the layers of compliance, it becomes burdensome.

Small and medium enterprises are different.

Decision-makers are close to the field. Therefore, it is easier to create a situation where everyone understands “What is this AI doing, and where does human judgment come into play?”

In the case of ¥50,000 script automation, the logic of processing is simple. “Fetch this data, sort it under these conditions, and output it in this format”—even a president who is not IT-savvy can understand this.

Because it is highly transparent, it is trusted. Because it is trusted, it is used. Because it is used, results are produced.

This virtuous cycle can be achieved precisely because the organization is a compact small or medium enterprise. Large companies cannot replicate this.

—

So, What Should Be Done?

The steps for small and medium enterprises to achieve results with AI utilization are actually simple.

Step 1: List the “tasks you do every day.” Transcription, aggregation, verification, notifications. Start by listing these. If there are routine tasks taking more than 30 minutes, those are candidates for automation.

Step 2: Consider whether it can be automated for under ¥50,000. Google Apps Script, Python, Zapier, Make (formerly Integromat). There are plenty of tool options available. Even if outsourced, many cases can be developed for ¥100,000 to ¥300,000 each.

Step 3: Use the time saved from automation for “decision-making.” This is the most important part. Simply reducing work hours will only lead to cost savings. Use the time saved to improve customer service quality, focus on new business development, or enhance products—the true return of automation lies in the reallocation of time.

Step 4: Once the structure of decision-making becomes clear, move to the next level of granularity. As small automations accumulate, there will be moments of realization, such as “This decision can be patterned.” At that timing, it is appropriate to consider advanced automation akin to AI agents.

There’s nothing wrong with a ¥3 million AI agent. It’s just a matter of the order being reversed.

—

Summary: Start Small and Uncover the Structure

¥50,000 automation and ¥3 million AI agents are not opposing choices. They represent different stages.

However, based on my experience observing small and medium enterprises, I can assert: 90% of companies that jump straight to ¥3 million will fail. Without understanding the structure of personalization, introducing advanced AI will only lead to more unused systems.

Automate tasks for ¥50,000. Use the time saved for human thinking. Through this thinking, the structure of operations will become visible. Once the structure is clear, move on to the next automation.

The companies that can sustain this unglamorous cycle will ultimately be the strongest.

The structure that allows small and medium enterprises to compete with large companies in the age of AI lies here. They can make quick decisions, maintain close ties with the field, and rapidly iterate small experiments. ¥50,000 automation is the first step.

Before purchasing expensive tools, I hope you will first list the tasks you will be doing tomorrow morning. The answer lies there.