Generative AI may feel like it’s been around forever, but we’re just getting started.
It’s been a little less than three years since Chat-GPT arrived. We’re now at the point where businesses must sharpen their strategies and move beyond first-generation adoption. The most pressing challenge now isn’t adding tools or even deploying them — it’s developing observability, sophistication, and strategy to capture what can feel like very distributed AI value.
Estimates vary but larger businesses are spending anywhere from $600-$1,400 per employee on AI tools this year. Yet 80% report no significant bottom-line impact from generative AI use.
This isn’t a technology failure—it’s a measurement failure. More specifically, it’s a sign that organizations are still struggling to see the value that AI is creating.
The problem is not, in fact, that AI doesn’t create value. It’s that most AI usage is invisible to the systems designed to measure that value, and a sound strategy going forward creates ways to for your business to observe, measure and take advantage of AI’s potential.
The Visibility Gap
McKinsey’s latest State of AI report shows 78% of organizations use AI in at least one business function. There’s a different struggle, though, for many businesses that many larger enterprises and Fortune 500 companies encounter: even more AI usage (by some estimates the overwhelming majority) happens outside of IT visibility.
Employees often use personal accounts for AI tools and features are quietly embedded in existing SaaS platforms. New tools are discovered and shared organically across teams. And IT sees very little of it, if any of it.
The result is a massive gap between what leadership thinks is happening and what actually is happening. This organic, distributed adoption represents AI maturation in action, but it challenges traditional enterprise management approaches and what company information remains private or is being fed into AI training models.
One healthcare system I encountered, for instance, approved 5 AI tools — but discovered 34 in active use. When you dig deeper, this isn’t an outlier: it’s an indicator of organizations transitioning from centralized AI experimentation to decentralized AI integration.
