82% of AI Engineering Spending Never Reaches Products—How to Solve the ‘80% Investment Evaporation’ Problem for Small Businesses with Just 50,000 Yen a Month

Investing 1 million yen in AI results in only 180,000 yen making it to market as a product. Where does the remaining 820,000 yen go?

This figure is derived from research indicating that only 18% of AI engineering spending actually reaches the market as a product. In other words, 80% of AI investments are ‘evaporating.’

This is the case for large corporations. It’s clear what would happen if small businesses were to engage in the same structure.

However, conversely, this presents an opportunity. If small businesses understand the structure that causes large corporations to lose 82%, they can invest in ways that avoid evaporation altogether. In this article, we will break down the ‘80% evaporation’ problem and explore how small businesses can derive tangible benefits starting with just 50,000 yen a month.

—

What’s Happening in Large Corporations—Uber and Starbucks’ ‘Expensive Tuition Fees’

First, let’s examine the structure of failures in large corporations.

Uber exhausted its AI budget in just four months. By allowing employees to freely use AI tools, usage skyrocketed. The costs of API calls far exceeded expectations, leading to the annual budget being surpassed in just four months. The issue was not with ‘using AI’ but rather that there was no system in place to track who was spending how much on what.

Starbucks introduced an AI agent but withdrew within months. They deployed an AI chatbot to improve customer service, but customer satisfaction did not increase, and in fact, it complicated on-site operations. This is a classic example of a situation where the definition of the problems AI should solve was vague, turning ‘implementing AI’ into an end goal in itself.

The common structure in these two cases is simple:

1. Weak problem identification (Unclear what needs to be solved)
2. Lack of cost management systems (Unlimited usage leads to budget depletion)
3. No metrics for measuring effectiveness (Unable to judge success or failure)

When these three factors come together, 80% of investments are sure to evaporate. Large corporations can withstand this due to their financial strength, but small businesses cannot. Therefore, it is essential to design a system that eliminates these three issues from the start.

—

Where Does the 82% Go?—Breaking Down the ‘Evaporation’ Breakdown

Even if it is said that ‘80% evaporates,’ if we do not know specifically where it is disappearing, we cannot take countermeasures. Let’s structurally break down the breakdown.

Looking at this table, the largest source of evaporation is ‘stopping at PoC.’ Prototyping is done, but it never makes it to production. This accounts for 30% of the total.

Next, the second largest is ‘over-investment in infrastructure and tools.’ There are many cases where a monthly AI platform costing tens of thousands of yen is contracted, but less than 10% of its features are utilized.

The lesson for small businesses is clear. If you are going to do PoC, design it from the start with the premise of going into production. Start with the minimum configuration of tools. Just these two steps can dramatically reduce the evaporation rate.

—

The Reversal Structure for Small Businesses—Why ‘Small’ Becomes a Weapon

Now, let’s change our perspective. The structure that causes 80% of AI investments to evaporate in large corporations is, in fact, a problem that arises because they are large corporations.

In large corporations, there are many stakeholders involved in AI implementation: management, IT departments, on-site staff, compliance, and external vendors. Decision-making takes three months, requirement definitions take three months, and development takes six months. During that time, the business environment changes, leading to ‘actually, it’s different’ situations. This is the true nature of the 17% return to the starting point.

What about small businesses?

• Decision-makers are close to the front lines (the CEO is observing the operations)
• Fewer stakeholders lead to quicker consensus
• A culture that allows for ‘let’s try first’ exists
• Even if failures occur, the impact is limited

In other words, a significant portion of the 82% that evaporates in large corporations can be structurally avoided. The distance between PoC and production is short. There are fewer returns to the starting point. Personalization can be easier to manage with a small team if systems are in place.

Utilizing AI in small businesses is not a scaled-down version of large corporations. It is an entirely different game that can operate at speeds and densities that large corporations cannot achieve.

—

Starting AI Investment with 50,000 Yen—Three Steps to Avoid ‘Evaporation’

So, what specifically should be done? Here’s a framework to minimize evaporation rates with a budget of 50,000 yen a month.

Step 1: Choose the ‘Most Painful Task’ (Investment: 0 yen, Duration: 1 week)

Before considering what AI can do, identify one task that currently incurs the highest cost, is the most personalized, or has the most mistakes.

For example:

• It takes two hours to create a quotation each time → 40 hours a month → Over 100,000 yen in labor costs
• Inquiry responses are concentrated on specific employees → If that person is absent, operations come to a halt
• Monthly report aggregation is done manually every month → One full day is lost each month

Start with ‘what is painful’ rather than ‘what can AI do.’ This alone significantly reduces the risk of stopping at PoC. This is because tasks that are painful have a high motivation for production deployment.

Step 2: Automate with Minimum Configuration (Investment: 30,000 to 50,000 yen a month, Duration: 2 to 4 weeks)

Attempt to automate the identified task using a combination of existing AI tools. Do not develop in-house.

Specific cost breakdown:

• ChatGPT Team: Approximately 4,000 yen/month per person (12,000 yen for 3 people)
• Zapier (automation tool): Approximately 3,000 yen/month
• Google Workspace: No additional cost if already in use
• Total: 15,000 to 50,000 yen/month

In the case of creating quotations, for example, generate drafts using ChatGPT based on past quotation data, and automate the retrieval of customer information and recording into Google Sheets with Zapier. Don’t aim for perfection. If something that used to take two hours can be reduced to 30 minutes, that alone saves 30 hours a month, equating to over 70,000 yen in labor costs. With an investment of 50,000 yen, you achieve a return of 70,000 yen. You start in the black from the first month.

Step 3: Systematize to Eliminate Personalization (Investment: Additional 0 yen, Duration: 2 weeks)

Once automation is successful, create a procedure manual. Record it in a video. Ensure that anyone can achieve the same results.

This is the most crucial and often overlooked step. If the responsible person leaves, the system cannot be operated—this is the essence of the 15% evaporation due to ‘personalization.’

The procedure manual doesn’t need to be elaborate. Just write in a Google Document, ‘1. Open this 2. Paste here 3. Press this button.’ Including screen captures is sufficient. The time required is 2 to 3 hours. This 2 to 3 hours can prevent future evaporation of hundreds of thousands of yen.

—

So, What Should We Do?

To summarize:

The reason 82% of AI investments evaporate is not a technical issue. It stems from starting with vague definitions of ‘what to solve,’ ‘how much to spend,’ and ‘how to measure results.’

Small businesses need to do just three things:

1. Select one painful task (not what AI can do, but what is painful)
2. Automate with existing tools for under 50,000 yen a month (do not develop in-house)
3. If successful, create a procedure manual to eliminate personalization (make it reproducible)

While large corporations spend millions and lose 82%, small businesses can derive tangible benefits with just 50,000 yen a month. They can make quick decisions, are closer to the front lines, and can ‘try first.’ This is not a weakness of small businesses but a structural strength.

Before signing contracts for expensive AI platforms, start by listing your company’s ‘most painful tasks.’ This will be the first step in preventing 82% from evaporating.