The Real Reason CEOs Are Worried About AI Skills
Nearly one in four CEOs say half their workforce will need AI reskilling within two years. But the problem isn't that employees aren't using AI. The real issue is that many are using it poorly, creating hidden risks instead of real value.
AI usage has become a dangerous vanity metric in business. Leaders track licenses, logins, active users, and prompts. These numbers look good on a dashboard. But they don't measure what matters: whether employees know how to use AI well.
What the Data Really Shows
The Conference Board's latest CEO Confidence survey reveals a startling finding: nearly 25% of CEOs expect to retrain more than half their staff for AI in the next two years. At first glance, this sounds like an adoption problem. But the truth is more complex.
In many companies, employees already use AI daily. They write with it, research with it, summarize documents, build slides, and analyze data. Yet, using AI and using it well are two very different things.
The ServiceNow Study: A Wake-Up Call
ServiceNow and Oxford Economics surveyed 4,500 executives for their 2025 Enterprise AI Maturity Index. They found that more than half had rolled out 100 or more AI use cases. But only 19% said those efforts drove meaningful business outcomes. The tools are everywhere. The activity is real. The value is missing.
Why Using AI Badly Is Worse Than Not Using It
AI doesn't help every task equally. A Harvard and Boston Consulting Group study found that consultants who used AI on tasks within its capability range performed much better. But on tasks outside that range, AI users were more likely to get wrong answers. The model gave them confident, polished, and incorrect results, and they ran with them.
Researchers call this the "jagged frontier." AI is brilliant on one side of a line you can't easily see and unreliable on the other. Knowing where that line sits for your work is the critical skill.
What Using AI Well Looks Like
- A salesperson who verifies three "facts" from an AI-generated customer brief before sharing it.
- An analyst who spots a clean-looking spreadsheet that's subtly wrong.
- A manager who knows which client data should never enter a public chatbot.
- A team that reworks a weak first draft instead of shipping it because the paragraphs sound professional but lack substance.
None of this shows up in a usage report. But it separates AI that boosts productivity from AI that creates expensive, invisible rework.
The Metric That Flatters Everyone
Most AI readiness is measured like gym memberships. Leaders count licensed seats, active users, and logins. These numbers feel like progress because they go up and to the right. But usage counts activity, not skill.
A team can run a thousand prompts a week and produce a thousand mediocre, unchecked, or wrong outputs. The adoption dashboard turns green, but the work doesn't improve. This is why so many AI initiatives stall in a fog of vague enthusiasm.
The Real Questions to Ask
- Do employees know when AI is likely to be useful?
- Do they know when AI is likely to be wrong?
- Do they verify outputs before using them?
- Do they know what should never be entered into a tool?
- Can they spot when a polished answer disguises weak thinking?
These skills don't appear in a usage report. But they determine whether AI becomes a productivity engine or a quiet source of risk.
The Leadership Blind Spot
Here's what should grab every executive's attention. ServiceNow found that the strongest predictors of AI value were not tools or budgets. They were leadership and governance.
Yet, in a Leadership IQ survey of 1,251 executives, nearly 80% said they personally use AI tools. But almost half either didn't believe AI would change their own role or weren't sure. Even leaders who use AI daily often lack a real understanding of what it means.
Most leaders can tell you how many employees have access to AI. They can tell you how many logged in last month. But they often cannot tell you:
- Which teams verify AI work before it reaches a client.
- Where employees paste sensitive information into unknown tools.
- Which departments redesign real workflows around AI versus just experimenting.
- Where unchecked AI output piles up, looking clean enough to pass.
That is the real reskilling problem.
How to Fix the AI Skill Gap
The next step for leaders who want AI to pay off is to change the question. Stop asking, "How many people use AI?" Start asking, "How well do they use it?"
Can your team tell when a model is likely wrong? Do they verify before they ship? Do they know what's unsafe to type into a public tool? Has anyone redesigned real work around AI, or is it still happening in scattered chat windows you never see?
These questions are harder to answer than a usage dashboard. That's exactly the point. They measure the thing that creates value instead of the thing that's easy to count.
The companies that pull ahead over the next two years won't be the ones whose people use AI the most. They'll be the ones whose people use it best, and whose leaders understood the difference early enough to do something about it.
Comment