AI Usage Up 52 Times, Cost Down 90%. Consider the Meaning of These Numbers

Normally, if you use something 52 times more, the cost would increase by 52 times. That’s obvious.

However, in the world of AI, this common sense has been turned upside down. One company increased its AI usage by 52 times and saw a 90% reduction in costs. In other words, a structure where “the more you use it, the cheaper it gets” has become a reality.

This is not just a technical issue. It’s about a fundamental change in the cost structure.

And here’s the important part: this structure is not exclusive to large corporations. In fact, small and medium-sized enterprises (SMEs) are in a better position to benefit from this inverted structure. Why? Large companies have existing systems and organizational structures that are cumbersome and slow to adapt. In contrast, SMEs can make quick decisions and start small, rapidly scaling up across the entire company.

The question is whether they are aware of this structure and whether they can act on that awareness.

Why Does “The More You Use, the Cheaper It Gets” Work?—The Essence of the Inverted Structure

First, let’s break down the mechanism behind the “52 times usage and 90% reduction”.

The cost structure of AI has fundamentally different characteristics compared to traditional business costs.

1. Marginal Costs Approach Zero

In traditional operations, as the workload increases, labor costs also increase proportionally. If one person is needed to process 100 invoices per month, then 10 people are needed for 1,000 invoices.

AI is different. Whether processing one item or 1,000, the additional cost is only the API usage fee. Moreover, that unit price is rapidly decreasing. OpenAI’s GPT-4 has seen its API price drop by over 90% from its release in March 2023 to the end of 2024. Google’s Gemini and Anthropic’s Claude are also engaged in similar price competition.

In other words, costs decrease just by waiting, but they decrease even further with increased usage.

2. The “Compound Effect” of Automation

When one task is automated with AI, it becomes easier to automate related tasks. For example, automating customer inquiries with an AI chatbot generates log data. Analyzing that data reveals patterns in frequently asked questions, enabling automatic FAQ generation. As the FAQ improves, the number of inquiries decreases.

When this chain reaction occurs, efficiency increases with usage, resulting in lower costs. This is the “compound effect” where one automation leads to the next.

3. Transformation of Labor Cost Structure

This is the most significant factor. In Japanese SMEs, the majority of costs are labor-related. If an administrative staff member earning 250,000 yen per month can be replaced by AI for just a few thousand yen per month, what happens?

Let’s look at a specific example. In a local manufacturing company, administrative staff spent three hours a day processing order emails. This amounted to approximately 1.2 million yen in annual labor costs. After automating this with AI, the monthly API cost was about 8,000 yen, totaling around 100,000 yen annually. From 1.2 million yen to 100,000 yen.

This is just one task. If the same applies to five or ten tasks, the savings could reach several million yen. And while usage may increase dozens of times, costs continue to decline.

The Bottleneck Is Not AI, But Humans

After reading this, you might think, “Let’s do it right away,” but the reality is not that simple.

The biggest bottleneck is not the performance of AI. It lies on the human side.

Let’s be specific.

Bottleneck 1: “Not Knowing What to Automate”

This is the most common issue. When asked, “What do you want to do with AI?” many SME owners respond, “Hmm, it can do something amazing, right?”

The problem is in the framing. Instead of asking what AI can do, the first step is to inventory “where time and money are currently being spent.”

How many hours per month does each person spend on what tasks? What is the hourly cost of those tasks? Just listing this information will reveal which tasks should be automated.

Bottleneck 2: “Unable to Act Due to Perfectionism”

Some business owners delay AI adoption because they expect 100% accuracy. They think, “What if there are mistakes?”

But let me ask: Is the current human work 100% accurate? Even veteran clerks make mistakes. If AI has 95% accuracy, combining it with human checks can bring it to over 99%. And the cost is far lower than doing everything manually.

If you wait for 100%, you will never start. Start at 95% and improve accuracy while operating. That’s the right approach.

Bottleneck 3: “Personalized Tasks Become Barriers”

“Only Tanaka understands this task”—this is a common scenario in SMEs. Know-how that exists only in Tanaka’s mind cannot be transferred to AI or passed on to others.

However, conversely, the introduction of AI presents an opportunity to eliminate personalization. To teach AI about Tanaka’s tasks, you need to create manuals. In the process, you may discover unnecessary steps or more efficient methods. AI can become a lever to break down personalization.

Three Concrete Steps for SMEs to Reproduce the “Inverted Structure”

Let’s move from abstract discussions to specific actions.

Step 1: Inventory Costs (1 Week)

First, list the internal tasks in terms of “time × labor cost.” Excel is sufficient.

Just creating this table will visualize “where the money is being spent.” In the above example, that’s 300,000 yen per month, or 3.6 million yen annually.

Step 2: Automate Small Tasks (1 Month)

From the inventory, choose the most routine task with minimal impact from errors.

Recommended tasks include “creating daily reports” or “responding to standard emails.” Failures in these tasks won’t be catastrophic.

Here are some specific tools and cost estimates:

• ChatGPT API (GPT-4o mini): $0.15 per input and $0.60 per output for 1 million tokens. Processing 100 daily reports a month costs only a few hundred yen.
• Google Apps Script + ChatGPT API: Automating email processing can be done for free by integrating with Gmail and spreadsheets.
• Zapier / Make (formerly Integromat): Automate workflows without coding for about 2,000 to 5,000 yen per month.

Initial investment is a few tens of thousands of yen, with monthly running costs in the thousands. A task that used to cost 3 million yen can now be managed for 50,000 to 100,000 yen annually.

Step 3: Expand Success (3 Months or More)

Once you achieve results in one task, share the numbers within the company. “Daily report creation reduced from 50 hours to 5 hours a month, saving 90,000 yen”—this concrete figure accelerates internal consensus for the next automation.

As you expand automation to the second and third tasks, the know-how for utilizing AI accumulates within the company. Initially, external support may be needed, but by the third task, you can manage it internally.

During this “expansion” phase, usage will increase significantly. However, costs won’t increase proportionally. This is because the design cost of the system is incurred only once, and subsequent implementations can reuse templates.

This is where the structure of “usage up 52 times, cost down 90%” is reproduced.

Don’t Imitate Large Corporations

One caution: many SMEs try to directly imitate the AI implementation cases of large corporations, but this is dangerous.

Large companies spend tens of millions of yen on proof of concept (PoC), form specialized teams, and take a year to implement. If SMEs try to replicate this, they will run out of funds at the PoC stage.

The strength of SMEs lies in “starting small, moving fast, and scaling across the company.”

• Instead of taking six months for a PoC, start with one task next week.
• Instead of forming an AI specialist team, have the on-site staff automate their own tasks.
• Instead of devising a company-wide strategy, expand from tasks that yield results.

With this approach, you can start with zero to a few tens of thousands of yen in initial investment. It won’t hurt if you fail. And if you succeed, you can achieve in three months what takes large corporations a year.

The Difference Lies in Whether You Act Now

AI costs are decreasing month by month. In 2024 alone, the API prices of major LLMs have dropped to less than half. This trend will continue into 2025.

In other words, the sooner a company starts, the longer it can enjoy the benefits of cost reduction. The longer you wait, the greater the gap will be with competitors who have acted first.

“It’s still too early for us”—if you hear this phrase, it’s a warning sign. A competitor in the neighboring town might be using AI for 50,000 yen a month while your company is spending 500,000 yen on manual labor. That day is closer than you think.

Start with one task. Next week. Open that Excel for inventory.