92% of Companies Are Betting on AI. Only 1% Know What They're Doing.

Nearly every company is doubling down on AI. Almost none of them are ready for it.

McKinsey's 2025 AI in the Workplace report puts the gap in plain terms: 92% of organizations plan to increase AI investment over the next three years. Only 1% describe themselves as mature in how they deploy it. That's not a technology problem. That's an execution problem.

The productivity numbers are real, but they come with a catch. Employees using AI tools on specific tasks like drafting documents, pulling data and writing code are reporting 20–30% efficiency gains. That's meaningful. The problem is those gains rarely show up in firm-wide output without deliberate workflow redesign. Individual wins don't automatically become organizational wins. Someone still has to do the plumbing.

For HR and benefits teams, this matters more than most departments want to admit. A lot of the work that buries admins...summarizing plan documents, answering routine enrollment questions, scheduling and basic analysis falls squarely in what McKinsey estimates as 60–70% of employee time currently spent on tasks generative AI can handle. That's not a threat. That's a description of work that shouldn't take this long.

But here's where the report gets honest in a way that a lot of AI vendors won't be: 65% of organizations have piloted AI. Fewer than 25% have scaled past the pilot stage. McKinsey calls this the capability trap. You run a proof of concept, it works well enough, and then it stalls because you don't have the internal talent to build on it, because people weren't brought along on the change or because no one can clearly measure what success looks like.

Benefits administration is a regulated, high-trust environment. Skepticism about AI here isn't paranoia and it's appropriate. Errors in this space have real consequences for real people. The capability trap is especially dangerous when teams feel pressure to adopt tools quickly without thinking through what happens when the tool gets something wrong.

The practical read for HR and benefits professionals: the question isn't whether to use AI. The question is whether your organization is doing the unsexy work, defining the right use cases, redesigning the workflows, measuring the right things that actually makes AI useful rather than just present.

Most companies are buying shovels and leaving them in the shed.

The 1% who describe themselves as mature aren't smarter. They just started doing the unglamorous work earlier. It's worth asking what that work looks like for your team specifically and whether the tools you're being handed were actually built for the job you're trying to do.