Most companies are using AI. Far fewer are getting lasting value from it. A new study from IDC, commissioned by Microsoft, surveyed more than 5,200 senior business and IT leaders across 28 industries to understand why that gap exists and what the most successful organisations are doing differently.

The numbers behind the gap

The study identifies a group called Frontier Firms: the top 10% of organisations on IDC’s AI maturity scale, which measures things like how broadly a company uses AI, how well it manages risk, and whether it has a plan to train its workforce. These firms report an average 4x return on their AI investments. Slow adopters, the bottom 30% on the same scale, report 2.7x. Among Frontier Firms, 39% see returns of 5x or more. Among slow adopters, only 4% do.

The gaps get wider when you look at specifics. Only 61% of all organisations are highly confident they can measure whether AI is actually delivering value. Among Frontier Firms, that figure is 93%. On skills, the divide is stark: 80% of Frontier Firms have an organisation-wide AI training strategy, compared with just 3% of slow adopters.

Meanwhile, 66% of leaders say AI has delivered substantial results, but only 48% are spreading those results across multiple parts of their business. The technology is working in places. It just is not spreading.

What the high performers do differently

Frontier Firms share five consistent habits. They spend more than they planned on AI: 90% exceeded their projected AI budgets. They use AI broadly across most of the business rather than in isolated pockets. They deploy agents, which are AI systems that can carry out tasks and make decisions with some independence, at scale: 96% are rolling agents out widely and 53% connect them across different workflows. They govern those agents carefully, with 87% having a formal oversight process, versus 51% of slow adopters. Finally, they invest heavily in people.

That last point is where the most striking gap sits. Most organisations report that their employees trust AI. Only 20% have a comprehensive plan to actually build AI skills across the workforce. Trust is running ahead of preparation.

Bristol Myers Squibb offers a concrete example. The company uses an agent called RABIT to help project managers track decisions and reduce follow-up work. Project managers who use it save an estimated 20% of their time, according to the study.

Conagra Brands approached it from the other end, starting with governance. Its Chief Information Officer, Tracy Schaefer, described the process:

“Laying the foundation of proper governance, observability and responsible AI was our first step to engaging our employees with AI.”

KOHLER’s Chief Digital Officer, Eric Tabor, made a point about the type of training most companies miss: not just how to use AI tools, but how to break down business processes and rebuild them as AI workflows.

What this means for people at work

The study’s most human finding is about the gap between trust and support. Employees broadly trust the AI tools they are using. Their employers, in most cases, have not given them a structured way to build real skills with those tools. That is a problem that will compound as AI takes on more tasks.

The good news is that the ceiling for what AI can free people up to do looks high. Leaders in the study pointed to creativity and innovation, customer relationships, and analytical thinking as the areas where people could do more if routine work is handled by agents.

What to watch

IDC Executive Analyst Shari Lava is leading sessions on AI return on investment at Microsoft Ignite on November 18, which may surface more detail from the research. With 71% of organisations expecting their AI budgets to grow by 10% or more over the next 12 months, watch whether companies start closing the skills gap or simply spend more without the workforce strategy to match.