Big Corporations Are “Backtracking” on AI. This Is Not Just Someone Else’s Problem.

Microsoft has eliminated a range of AI features. Lululemon’s AI chief left after less than a year. Meta’s AI model “Glimmer” is facing criticism over data ethics.

Even large corporations with investments in the hundreds of billions are stumbling badly with AI.

It’s easy to dismiss this as merely a “big corporation issue.” However, there are structural lessons here that small businesses must confront. By understanding where large companies have stepped on landmines, small businesses can avoid those pitfalls. Moreover, these landmines are often easier for small businesses to sidestep.

—

Landmine #1: Microsoft’s “Everything Syndrome”—The More Features, the Less Value

Microsoft has discontinued AI-generated podcasts, group chats, deep research, and Mico characters. All of these were technically feasible.

The problem was that it was unclear who, when, and why would use them.

Microsoft has shifted its strategy to integrate these features into its Copilot app, consolidating consumer and business offerings into a single “super app.” In other words, they are cleaning up the mess they created. Reports suggest that the company’s AI-related investments could reach $80 billion (approximately 12 trillion yen) just for the fiscal year 2024, with a portion of that disappearing into “unused features.”

The lesson for small businesses is simple.

Start not with “what can AI do” but with “what is currently painful in the field.”

For example, creating estimates that take 30 hours a month. Spending 5 hours a week responding to inquiries. By applying AI to these tasks, the impact can be measured, and even if it fails, the damage will be minimal.

Microsoft failed by trying to do “everything.” Small businesses can start with “just this.” Having fewer resources actually forces better decision-making. Automate one task with an AI tool costing 2,000 to 5,000 yen per month. If that saves 20 hours a month, that translates to a value of 40,000 to 100,000 yen based on hourly rates. The return on investment is clear.

—

Landmine #2: Lululemon’s “AI Chief Exits After One Year”—Without Organizational Change, Technology Won’t Stick

Lululemon hired a head of AI, who left less than a year later. While the specific reasons have not been disclosed, there is a pattern to such stories.

The illusion that “hiring AI talent will bring change.”

Even if you hire excellent AI talent, if the management does not understand the essence of AI, a gap will form between them and the field. Vague directives like “increase sales with AI” and resistance from the field to change existing workflows, coupled with a lack of a well-organized data foundation, leave the AI chief caught in the middle, unable to deliver results, and ultimately resigning.

This occurs in large corporations because the cost of “changing” is too high due to their size. There are too many elements to coordinate, such as inter-departmental adjustments, approvals, and relationships with existing vendors.

This is where small businesses have an advantage.

If the CEO decides to “do it,” changes can be made as soon as next week. No approvals or inter-departmental coordination are needed. In a company of ten, one meeting can align everyone in the same direction.

What can be done concretely?

• Hire an external AI expert for a monthly “brainstorming session” (around 30,000 to 50,000 yen).
• Assign existing members as “business improvement officers” rather than “AI officers.”
• Focus on “automating one task with one tool” for the first three months.

Lululemon thought that hiring an expert would solve the problem. Small businesses can move faster by leveraging their own knowledge of their operations to master the tools. The subject of AI utilization is not “AI talent” but “the people on the ground.”

—

Landmine #3: Meta’s “Glimmer Issue”—Ignoring Data Handling Can Destroy Trust

Meta’s open AI model “Glimmer” is technically noteworthy. However, it has faced a barrage of criticism regarding the transparency of its training data and ethical considerations. Questions about which data was used for training and whether user consent was obtained remain inadequately addressed.

For Meta, this may just be another “backlash.” Large platforms have the capacity to absorb such controversies.

But small businesses do not have that capacity.

Feeding customer data into AI tools for analysis is convenient. But where is that data stored? Is it sent to external AI services? Has customer consent been obtained?

For local small businesses, trust is their most significant asset. Once a reputation arises that “that company mishandles customer data,” it can be devastating for business in the community.

There are only three things to do.

1. Read the terms of use for AI tools. You can check in five minutes whether customer data will be used for training. If using ChatGPT via API, the data will not be used for training.
2. Establish operational rules to avoid directly entering customer data. Simple rules like anonymizing names or masking identifiers are sufficient.
3. Be prepared to explain to customers, “This is how we use AI.” Just being able to answer when asked can help maintain trust.

The cost is zero. What is needed is only “awareness.”

—

So, How Should Small Businesses Move Forward?

While I have outlined three failures of large corporations, the underlying structures are common.

• Microsoft’s failure: Adding features without purpose → Start with “what to solve”
• Lululemon’s failure: Relying on talent for implementation → Operate with “the field as the subject”
• Meta’s failure: Underestimating data ethics → Create “rules to protect trust” first

All of these are easier for small businesses to address. They can make decisions quickly, the field and management are closely aligned, and they have a tangible understanding of the weight of the trust they need to protect.

Here’s the first step you can take starting tomorrow.

1. Choose one “routine task that takes more than 10 hours a month” within the company.
2. Test one AI tool that could be useful for that task (a free trial is sufficient).
3. Use it for two weeks and record whether “time decreased” or “quality changed.”

With a monthly cost of a few thousand yen and almost zero risk, saving 10 hours a month translates to 120 hours a year. Even at an hourly rate of 2,000 yen, that generates a value of 240,000 yen.

Small businesses can avoid landmines that large corporations have stumbled over at the cost of billions, for just a few thousand yen. This is the greatest advantage for small businesses in AI utilization by 2025.

—

Future Points of Interest

It will be crucial to observe how Microsoft’s Copilot integration unfolds. If the “super app” succeeds, it will change the options for AI tools available to small businesses. Additionally, as the EU’s AI regulatory law (AI Act) comes into full effect in late 2025, rules regarding data ethics will become even stricter. By summarizing “your company’s AI usage rules” on a single sheet of paper now, you can significantly reduce future compliance costs.

The failure news from large corporations serves as excellent educational material for small businesses. They can learn for free what large companies have proven to be “what not to do” at the cost of hundreds of billions.