Title

Despite Rapid Growth of AI, Productivity Remains Stagnant—Three Conditions for SMEs to Break Free from the ‘No Change After Implementing AI’ Mindset

Body

“We have implemented AI”—but how much has sales increased?

Very few small and medium-sized enterprises (SMEs) can answer this question immediately. That is the reality.

It has been two and a half years since the emergence of ChatGPT. The number of AI tools has exploded, and it is now possible to use decent ones for a monthly fee of around 20,000 yen. Nevertheless, looking at productivity statistics in the United States, there is no clear correlation between the spread of AI and productivity improvement. This is even more pronounced in the context of Japanese SMEs. “Nothing has changed even though we implemented AI”—this sentiment is no longer rare.

Why does this happen? And how can we break free from it? Let’s organize this from a structural perspective.

“Implementing AI” and “Changing with AI” Are Completely Different Matters

First, let’s confront the common failure patterns.

A typical flow in the AI implementation process for SMEs is as follows:

1. The president declares, “We need to adopt AI too.”
2. They sign up for a corporate contract with ChatGPT.
3. Some employees try it out but leave it untouched because they don’t know how to write prompts.
4. Three months later, only the monthly fees remain.

This is not “implementation”; it is merely “subscription.”

There is a huge gap between introducing a tool and changing business operations. It is structurally the same as a company that has Excel installed but still manages records on paper.

Three Structural Reasons Why Productivity Does Not Increase

So, why doesn’t productivity increase even after implementing AI? Before looking at individual cases, we need to understand the structure.

Reason 1: Costs Saved in One Area Are Consumed by Costs in Another

AI has reduced the time for document creation to one-tenth. That’s fantastic. But what if checking that document still takes the same amount of time as before? What if a supervisor spends an hour reviewing an AI-generated proposal, asking, “Is this really okay?”

When viewed in total, there are many cases where the overall workload has not decreased. This is a classic example of local optimization not leading to overall optimization.

The case of legal AI “Harvey” is emblematic. According to reports, Harvey’s profit margin plummeted from +50% to -50% in just six months. While AI increased the speed of legal document creation, the surge in quality control costs, API usage fees, and the complexity of customer interactions hit all at once.

This is not just a problem for Harvey. The phenomenon of bottlenecks merely shifting before and after the “faster process with AI” can occur in any industry.

Reason 2: There Is No Clear Purpose for Use

Starting with “What can AI do?” almost always leads to failure.

The correct question is, “What are we struggling with right now?” Processing invoices at the end of the month takes three days every time. Compiling sales reports takes two hours twice a week. The document screening process for hiring is so personalized that only the president can do it.

Without specific “pain points” like these, implementing AI will lead to a lack of use cases. Naturally, there will be no results.

Reason 3: The Business Processes Themselves Are Not Changing

This is the most deep-rooted issue. Most companies are using AI merely as a tool to speed up existing operations.

For example, using AI to create estimates. What used to take 30 minutes is now reduced to 5 minutes. A 25-minute reduction. Not bad. But the five-step process of hearing, internal confirmation, supervisor approval, creation, and sending the estimate remains unchanged.

To truly increase productivity, the process needs to be redesigned to the extent that “AI automatically structures the hearing content, generates the optimal estimate from past similar cases, notifies the approver via Slack, and sends it with one click.” Instead of reducing 30 minutes to 5 minutes, we need to aim for a state where the 30 minutes is “automatically completed.”

Whether or not we can delve this deeply is the dividing line for results.

What Sets Successful Companies Apart—Three Conditions

So, what are the successful companies doing? I will distill the common points I have seen in the field into three conditions.

Condition 1: Focus on “One Task” and Measure Effectiveness with Numbers

Successful companies do not start with a company-wide implementation. They begin by deciding, “We will improve this task, with this KPI, by this deadline.”

The case of Proaction in the United States is instructive. By utilizing OpenAI’s Codex and concentrating their efforts on specific business processes, they achieved a 60% increase in sales and a reduction of over 75 hours in operational time. They achieved results by targeting pinpoint areas rather than broadly saying, “Let’s utilize AI company-wide.”

For SMEs, starting with just one task is sufficient. “Reduce month-end invoice processing from three days to half a day” or “Automate the compilation of sales reports to reclaim two hours a week.” With specific goals like these, effectiveness can be measured, and success stories can emerge within the company.

Condition 2: Create a “Method of Use” Rather Than Relying on “Knowledgeable People”

“A young employee who is knowledgeable about AI is managing it alone”—this is personalization, not a system.

If that young employee leaves, it’s over. This is a common sight in SMEs, but it does not scale.

Successful companies create a method that specifies, “Use this prompt for this task, output in this format, and verify with this checklist.” It can be called a manual. It is a system that operates at the same quality and speed, regardless of who performs it.

Training and education are important, but even more crucial is creating a state where it can be used without thinking. It’s like a cooking recipe. Instead of relying on the chef’s intuition, we aim for a state where following the recipe yields an 80-point dish.

Condition 3: Reflect Weekly and Change Monthly

The speed of evolution of AI tools is extraordinary. What was considered best practice three months ago can become outdated today.

Therefore, the worst choice is to “implement and be done.” Successful companies reflect weekly on “How many hours did we save with AI this week?” and “Where did we stumble?” and update their operational rules monthly.

This is difficult for large companies to replicate. They have to go through approvals, implement company-wide, and revise manuals—large companies cannot keep pace with the evolution of AI at that speed. It is precisely because SMEs can make decisions quickly that they can “try something last week and change it this week.” This is the greatest weapon of SMEs.

So, What Should We Do?

To summarize:

• Do not implement company-wide. Focus on one task.
• Do not rely on knowledgeable individuals. Create a method.
• Do not just implement. Reflect weekly.

The costs of AI have dramatically decreased. Models comparable to GPT-4 can be used for a few thousand yen per month. Image generation, voice recognition, and translation, which would have cost hundreds of thousands of yen to outsource two years ago, can now be handled with tools costing 10,000 to 20,000 yen per month.

In other words, whether or not you can use AI is no longer a differentiator. What sets companies apart is whether they can change the flow of operations with AI.

For SMEs, this is nothing but an opportunity. While large companies struggle to move due to organizational weight, SMEs can make the judgment to “change one thing first” on the ground.

The reason for “no change after implementing AI” is not due to AI itself; it is a matter of how to change.

First, I want you to choose one of the most troublesome tasks in your company tomorrow. If you start from there, the scenery will change.