A new survey from the consulting firm BCG, covering more than 1,300 technology leaders, found something that should worry anyone tracking company budgets: 80% of AI spending now sits outside the IT department. The finance team, the marketing team, operations, HR, each is buying and running its own AI tools, often with no one in the company adding up the full bill.
The money involved is growing fast. AI spending went from under 2% of company revenue in late 2025 to more than 3% now, in less than a year. That is not a rounding error. For a mid-sized company, that kind of jump can mean millions of extra dollars a year, scattered across tools that nobody is checking for results.
This spending spread is not necessarily a mistake. AI is being used to change how entire departments work, not just how the IT team runs servers, so it makes sense that budgets would follow the work. The problem is that most companies have no one whose job is to look across all of it and ask a simple question: is this actually paying off?
That question is getting harder to answer honestly. A widely cited MIT study found that 95% of company AI projects produce no measurable financial return. Separately, Accenture found that while employees say AI is making their work more productive, the leadership teams funding it are struggling to point to real business results. The gap between what AI promises and what it delivers is wide, and it is not closing on its own.
BCG's advice cuts against how most companies have been spending. Their research found that only 10% of what makes an AI project succeed comes from the algorithm itself, another 20% comes from the data and technical setup, and a full 70% comes from changing how people actually do their jobs, redesigning workflows, retraining staff and rebuilding who is responsible for what. Most companies are doing the opposite: buying expensive software and leaving the actual work untouched.
The newer wave of AI, often called agentic AI, raises the stakes further. Unlike older tools that just answer a question or draft a document, these systems are built to carry out a full chain of tasks on their own, the kind of work that used to require a person checking each step. Companies are starting to report real value from this, but only when someone has set up clear rules for who owns the outcome if the AI gets something wrong.
There is a personal cost to getting this wrong. A recent survey found that 61% of tech leaders fear losing their job if they cannot show their AI investments are working. That pressure will not stay confined to the IT department. If 80% of the spending is happening in other departments, the blame for a failed AI bet is headed there too.
For a company outside the tech industry, the lesson is not to slow down on AI. It is to stop treating it as a collection of separate purchases. Someone needs to see the full list of AI tools bought across every department, what each one cost, and what business result it was supposed to produce. Without that, a company cannot tell the difference between a tool that is quietly saving money and one that is quietly burning it.