Ask a business owner why they adopted AI and the answer is almost always about time. Faster drafts, faster reports, faster everything. Ask the same owner what it’s actually costing them and the answer gets a lot less confident.
New research from Asana, published in June 2026, found that 82% of IT leaders experienced unplanned or unexpected AI-related cost increases in the past year. Not because the subscription went up. Because nobody budgeted for deployment, integration, governance, and the dozen small costs that show up once a pilot becomes something the whole business actually runs on.
The invoice isn’t the expensive part
Here’s the number that should worry you more. 53% of IT leaders say an AI tool or agent took an action in the past year that landed them in real trouble: financial damage, legal exposure, or a hit to reputation. That’s not a subscription line item. That’s a mistake nobody caught before it went out the door.
Meanwhile, 61% of decision makers say they’re fully accountable for what AI does in their business. Somebody signs off on the tool, and somebody else absorbs the consequences when it acts without the full picture.
The time savings aren’t automatic either
Almost half of failed AI projects, 46%, fail for the same reason: the tool never had the context it needed. It didn’t know your workflows, your internal knowledge, or how your business actually operates. It just guessed, confidently, and got it wrong in a way that looked right.
That has a real cost too. More than a third of workers, 37%, spend upwards of 30 minutes every day correcting AI output because it lacked context. Add that up across a team and the “time saved” story starts to look thinner than the demo suggested.
Tool sprawl makes it worse
One department picks up a content tool. Someone else finds a research assistant. A third person starts using a personal AI account because the approved option doesn’t do what they need. A quarter of workers, 25%, say they regularly use AI tools their company hasn’t approved, and 38% use personal accounts for work tasks.
Now you’re paying for tools nobody’s tracking, while your actual work data moves through systems your business never vetted. That’s not a productivity story anymore. That’s a budget problem wearing a productivity costume.
What actually works
Stop asking which AI tool to buy and start asking where the work is actually slowing down. A specific bottleneck, customer replies taking too long, a report that eats four hours a week, gives you something to measure. “Everyone else is doing it” doesn’t.
Before you roll a tool out further, give it the context it needs: your process documents, your past examples, your actual workflow. And know which tools your team is really using, not just which ones you approved. That’s the difference between AI that pays for itself and AI that just shows up as a surprise on next quarter’s invoice.
If you want help figuring out where AI is actually saving you money versus where it’s quietly costing you more than the subscription suggests, our strategic IT guidance can help.