I recently spoke with a board member at a company that, on paper, had everything going for it on the AI front. Strong technical team. Relatively clean data. Real internal momentum. They'd launched dozens of AI pilots, many of them technically successful, and generated genuine enthusiasm across the organization.
And none of it moved the needle.
This is far more common than most leaders realize. The problem isn't execution. It's not talent, and it's not technology. The problem is that most organizations are solving AI problems at the wrong level of abstraction.
The Car Analogy
Imagine you drive an hour each way to meet someone. How does AI help?
Level One: AI builds better routing software and saves you ten minutes on the drive.
Level Two: AI drives the car for you, so you're productive for the full hour.
Level Three: You take the meeting on video, AI captures the notes and action items for you, and you eliminate the trip entirely.
All three are valid applications of AI. But they produce radically different amounts of value. And they require radically different people to identify them and make the call.
This plays out in enterprises every day. When a retailer's AI team builds a better demand forecasting model, that's Level One. When the CEO asks, "What if AI fundamentally changes how customers discover and buy our products?", that's Level Three. Same company, same technology, completely different value.
Why Most Pilots Get Stuck at Level One
In most organizations, the people most enthusiastic about AI are the junior employees. They're less constrained by organizational memory, they are learning fast, experimenting constantly, and eager to prove the technology works. That energy is invaluable.
But these teams tend to stay tactical for two reasons.
First, they lack the full context. The demand forecasting team often doesn't understand all the marketing, product, and merchandising levers that go into how customers find and discover products in that category.
Second, and more subtly, sometimes, they don't feel they have the intellectual license to think that broadly. Their job is to optimize the model, not reimagine the business. No one told them to think bigger, so they don't.
Meanwhile, the senior leaders who do have the strategic context (the board, the CEO, the executive team) are often the least engaged with AI's tactical capabilities. They know where the value is, but they're not close enough to the technology to connect it to the right problems.
This is a structural gap, not a people problem. And it's where pilot graveyards are built.
The Real Job of Leadership
Running pilots is necessary. You need to let the water flow through the pipes, build organizational muscle, and understand how the technology actually behaves in your environment. No one should skip that step.
But the step most companies are skipping is the one that matters most: senior leadership leaning in to identify where the real value lives. Not delegating "the AI thing" to a task force. Not waiting for bottom-up pilots to organically discover strategic opportunities. Actually doing the hard work of asking: At what level of abstraction should we be solving this problem? And then making appropriate bets.
This means three things in practice.
Provide context and license. Leadership must give teams both the strategic context and the explicit permission to think beyond pilot owners functional boundaries. If your AI team is only optimizing within the lines they were drawn, that's a leadership failure, not a team failure.
Create deliberate collisions. The best AI programs I've seen deliberately mix levels of hierarchy and functional groups when designing pilots. Put the data scientist in the room with a marketer and the operations lead. Put the junior engineer next to the board member. The collision between AI fluency and strategic context is where Level Three thinking emerges. It doesn't happen organically. You have to design for it.
Have courage. There is no substitute for courage. Leaders at the senior-most levels need to lean in, make bold bets, and create a culture where it is okay for other leaders to fail. Level Three answers are inherently riskier than Level One optimizations. If your organization punishes failure, it will never get past incremental improvements.
That question, "What is the right level of abstraction at which the problem can be best solved using AI", is the question that separates companies capturing real value from companies generating impressive demos.
The best AI strategy isn't about making Level One faster. It's about recognizing when the problem demands a Level Three answer and having the leadership conviction to pursue it.