Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

Here’s why your AI rollout is failing

Most leaders assume their workforce can figure out AI tools, but the data says otherwise.

[Photo: Getty Images]

Profile photo of Matt Alldian

Founding Partner · Cardboard Toast

Here’s a number that should terrify every executive who approved an AI budget this year: Only 14% of workers use GenAI daily.

That’s according to PwC’s 2025 survey of nearly 50,000 workers across 48 countries. Despite billions in enterprise AI investment, despite “intuitive” interfaces, despite vendor promises that “anyone can use it”—adoption has barely moved from 12% last year.

Meanwhile, 63% of employers tell the World Economic Forum that skills gaps are their number one barrier to AI transformation. Not budget. Not technology. Skills gaps.

This isn’t a technology problem. It’s a training problem. And most organizations are solving it backwards.

The skills gap is worse than you think

Most leaders assume their workforce can figure out AI tools. The data says otherwise.

An OECD study of 215,942 adults across 33 developed countries found that only 5% have the high-level computer skills that AI use requires—the ability to define implicit goals, handle unexpected outcomes, and evaluate information reliability.

Another 26% can handle medium-complexity tasks. That leaves 69% of the workforce unable to complete anything beyond basic, single-step operations.

Here’s the critical point: This isn’t a capability problem—it’s a training problem.

These skills are learnable. The issue is that most organizations deploy AI assuming workers already have them, or that a two-hour workshop will bridge years of missing foundational development.

If you’re reading Fast Company, you’re almost certainly in that top 5% of adults with high-level computer skills. These skills feel natural to you. And that’s exactly the problem—you’re designing AI experiences for a workforce that hasn’t been trained to use them.

The National Skills Coalition found that 92% of jobs require digital skills, yet one-third of American workers have “little to no” foundational digital literacy. When you deploy AI assuming workers can “just figure it out,” you’re assuming training that was never provided.

Why prompt training fails

The typical employer response is to offer workers a prompt engineering workshop. This fundamentally misunderstands the problem.

Prompt engineering assumes workers can already define goals clearly, evaluate outputs, and iterate based on results. It focuses on “how to ask,” when the real gap is “how to evaluate what you get back.” You can’t teach prompting to someone who hasn’t learned to assess whether an output is accurate, complete, or hallucinated.

LinkedIn’s 2025 Workplace Learning Report found that only 36% of organizations have robust training programs. The rest have limited programs or nothing at all. You can’t close a training gap this large with occasional workshops—you need systematic capability building.

The trust problem

Here’s what makes this harder: Workers won’t use tools they don’t understand or trust.

PwC found that workers who feel supported to upskill are 73% more motivated than those who don’t. But only 50% of workers trust top management, and only 15% got help building a career plan this year—down five points from 2024.

Trust and training are inseparable. A workshop can’t train people to trust AI. Trust develops through understanding how the technology works, recognizing when outputs are reliable, and building confidence through structured practice.

This is why deploy-first approaches fail. You can’t build trust through deployment. You build trust through training, then deploy.

The proof training-first works

LinkedIn compared organizations with robust career development programs against those without. The results: Training-first companies are 42% more likely to be frontrunners in AI adoption. Fifty-one percent of them are at “leading” or “accelerating” stages in AI, compared to just 36% of organizations without strong training programs.

This also explains why productivity studies find wildly variable results—anywhere from 0% to 16% gains depending on the workflow. The high-gain cases either required minimal evaluation (filling gaps where nothing existed) or supported existing expertise. The failures? They required workers to trust AI outputs in domains where they lacked the training to evaluate quality.

That’s nearly a 1.5x advantage. The companies winning with AI aren’t those with the best models—they’re those that invested in training their people first.

What leaders should do

Here are five actions leaders can take to bridge the training-trust gap.

1. Assess before deploying. Stop assuming skills exist. Before rolling out AI tools, understand your workforce’s actual computational literacy—not what you hope they have.

2. Build foundational skills first. Train workers on evaluation, goal articulation, and iteration before you train them on prompting. These skills transfer across every AI tool.

3. Develop trust calibration. Teach people when AI is reliable and when it isn’t. Workers need to know what hallucination looks like in their domain, when to verify outputs, and how to spot errors.

4. Start with assisted use. Deploy AI in human-in-the-loop workflows that support learning. Don’t start with autonomous agents—start with copilots. Autonomy comes after competence.

5. Measure trust, not just adoption. Track whether workers can spot bad outputs, know when to trust versus verify, and use AI for appropriate tasks—not just whether they’ve “tried it.”

    Final thoughts

    The World Economic Forum projects that 59% of workers will need training by 2030 and 39% of current skills will become obsolete. You can’t workshop your way through that gap. The 42% advantage goes to organizations willing to do the systematic work of building skills and trust before expecting results.

    The technology isn’t the bottleneck. The training is.

    Matt Alldian is a founding partner of Happy Robots.