Enterprise AI spending hit $37 billion in 2025—a 200% jump from the year before. The message from the C-suite couldn’t be clearer: AI is no longer a competitive advantage. It’s table stakes.
So why are three-quarters of enterprises still stuck in pilot mode?
Budgets have been approved, platforms deployed, and centers of excellence stood up. Yet few AI initiatives meet expectations for revenue impact. The technology isn’t the problem. The problem is that no one actually taught your people how to use it.
The knowledge gap is enormous
Enterprises are running an average of 200 AI tools. However, only 28% of employees know how to use their company’s applications, and only 7.5% have received what could be called extensive AI training. And when employees don’t get trained on sanctioned tools, they route around IT, quietly using whatever works, hidden from view.
The behavior that follows is predictable: 57% of American employees are reluctant to admit to their managers that they use AI at all. Nearly half admit to pretending they know how to use it just to avoid looking incompetent. Leaders, meanwhile, use AI at double the rate of individual contributors, meaning the people doing the most hands-on work are the least supported.
This isn’t a motivation problem. Employees want to use AI. They’re already using it. They just don’t know how to use it in ways that generate real business value, and most organizations haven’t given them a meaningful opportunity to learn.
Building AI fluency requires practice
Most enterprise AI training follows a familiar script: a video module, a PDF of prompt examples, maybe a vendor demo. Employees watch, nod, and return to their desks, where they proceed to use Claude or ChatGPT as a slightly smarter search engine.
The most durable professional skills, sales, negotiation, clinical procedures, public speaking, are learned through practice and real-time feedback, not passive consumption. AI fluency is no different. Yet the overwhelming majority of enterprise training isn’t built this way. We’re expecting behavior change from an experience designed to check a compliance box.
The result is predictable: 42% of companies abandoned most of their AI initiatives in 2025, up from 17% the year before. Not because the tools failed. Because the humans were never really brought along.
What experiential learning actually looks like
The organizations closing the adoption gap are treating AI fluency the way every other high-stakes skill gets built: through practice, feedback, and measurement. Flight simulators don’t exist because pilots lack motivation, they exist because repetition in a realistic environment is how competence actually forms. The same logic applies to AI. A video module watched once doesn’t change behavior. Guided practice, applied in context, does.
Leading organizations are acting on this. Some embed AI coaching directly into workflows so employees receive guidance at the moment of use. For example, Morgan Stanley built a GPT-powered assistant that helps financial advisors retrieve research and insights during client conversations.
Still others cultivate internal AI communities of practice, shared spaces where employees troubleshoot, trade prompts, and develop organizational norms around responsible use. For example, PwC has a firmwide AI upskilling initiative that has led to grassroots “prompting parties” and brainstorming sessions where employees experiment with generative AI and share use cases together. Google Cloud took a structured certification approach to train 15,000 sales reps on go-to-market strategy, producing not just trained employees but a measurable, consistent standard for how AI gets used across the field.
The data supports the investment: companies with a formal AI strategy report 80% adoption success, compared to 37% for those without one.
The zombie center of excellence problem
In 2021, zombie companies were created, businesses valued at billions with little revenue to show for it. I worry we’re now creating zombie centers of excellence inside enterprises: teams that have spent hundreds of millions on platforms that nobody uses in their day-to-day work.
The companies that win the AI era won’t be the ones with the biggest budgets. They’ll be the ones with the most skilled users. Mandating a tool doesn’t make people use it. Buying a license doesn’t create fluency. Real adoption requires meeting employees where they are, building confidence through practice, and measuring whether behavior is actually changing—not just whether someone completed a module.
The $37 billion question isn’t whether your company is investing in AI. It’s whether that investment is landing.