42% of companies abandon most AI projects (up from 17% last year). Why? They bought tools before defining problems, underestimated adoption costs, and never tracked baseline metrics. The fix isn't sexy: start small, measure obsessively, kill fast.
Here's a number that should terrify AI vendors: 42% of companies are now abandoning most of their AI projects. Last year it was 17%. That's not a trend. That's a collapse in confidence.
S&P Global published this data in early 2025. The AI industry mostly ignored it. Wonder why.
The Gap Between Promise and Reality
Let's talk about two numbers that don't add up.
That's not a rounding error. That's a 50x gap between what's promised and what's delivered. Companies aren't stupid. They're noticing.
And we're not talking about failed experiments here. These are tools companies paid for, trained people on, integrated into workflows. Real money. Now sitting unused.
Three Ways Companies Burn Money on AI
1. Solution Shopping
"We need AI" is not a business problem. It's FOMO dressed up as strategy.
Teams buy ChatGPT Enterprise because their competitor did. Then they spend three months trying to figure out what to do with it. Classic solution looking for a problem. And it almost never works.
Write down the exact task you want automated and how many hours per week it takes. If you can't, stop shopping.
2. Underestimating the Human Problem
The best AI tool in the world is worthless if nobody uses it. This sounds obvious. Companies still get it wrong constantly.
McKinsey broke down what actually determines AI project success. The technology itself? Only 25-30% of the equation. Leadership commitment and change management? 40-50%. Most of the budget goes to the smaller factor.
3. Flying Blind
If you don't know how long a task takes before AI, you can't prove AI made it faster. Shocking how many companies skip this step.
Then budget review comes around. "How much is this tool saving us?" Silence. Tool gets cut. Everyone acts surprised.
The Uncomfortable Truth About Timelines
Microsoft tracked Copilot users. Average time saved: 11 minutes per day. Not bad. But here's the part they mention quietly: it took 11 weeks for those savings to appear.
Eleven weeks. Most companies evaluate AI tools at 30 days. They're measuring during the worst part of the curve and calling it a failure.
Expect productivity to DROP in weeks 1-4. If you're not prepared for this, you'll kill a tool that would have worked.
What Actually Works
The companies seeing real returns aren't doing anything revolutionary. They're just being disciplined about basics most people skip:
They start with ONE task, not a company-wide rollout. They measure time before and after, obsessively. They budget real hours for training—not "watch this 5-minute video" but 10-20 hours per person. And they set a kill date. If it's not working by day 90, it's gone.
"AI's value depends on how well employees adopt the system. If adoption is low, ROI is lower, even if the AI itself is effective."
— IBM Institute for Business Value
The Honest Checklist
Before you buy anything, answer these. Honestly.
Can you name the exact task this tool will automate? Do you know how many hours that task takes right now? Have you calculated the break-even point in actual dollars? Who on your team will champion this and make sure people use it? What's your kill criteria—the specific point where you'll admit it's not working?
Can't answer all five? You're not ready. That's not an insult. Most people aren't ready. The difference is whether you figure that out before or after you've burned the budget.
For a deeper look at costs most buyers overlook, read our analysis of the hidden costs of AI tools. You can also check 5 signs an AI tool will actually save you time and our AI productivity research roundup for the real data behind the hype.
The TaskROI team researches AI productivity tools and helps businesses calculate real ROI before purchasing. Our data comes from industry studies by McKinsey, Harvard Business Review, and the Federal Reserve.