McKinsey’s State of AI survey finds 88 percent of organisations using AI in at least one function, 62 percent experimenting with agents and 23 percent scaling an agent anywhere. Within any single function, fewer than 10 percent are scaling agents.
Gartner predicts more than 40 percent of agentic AI projects will be cancelled by the end of 2027. S&P Global data, as reported in industry analyses, shows the share of enterprises abandoning most of their AI initiatives jumping from 17 percent in 2024 to 42 percent in 2025.
Spending on agentic AI keeps climbing anyway. Most stacks have a pilot somewhere with no owner, no target number and a renewal date next quarter. Sooner or later the CFO asks which side of the funnel it’s on.
| Category | Share of organizations |
|---|---|
| Using AI in ≥1 function | 88% |
| Experimenting with agents | 62% |
| Scaling agents anywhere | 23% |
| Scaling in a given function | 10% |
Cite or embed this
Why do pilots die?
Almost none of the post-mortems blame the AI. There are four causes.
No owned number. McKinsey’s high performers are the 5.5 percent who attribute more than 5 percent of EBIT to AI. They tied the work to a number someone owns.
No redesign. High performers are 2.8 times as likely to have fundamentally redesigned their workflows, 55 percent against 20 percent. Bolt an agent onto an unchanged process and you just automate its problems.
Counting costs per token instead of per project. That’s why Gartner lists escalating costs first. I break that bill down in the full bill for one agent.
No kill criterion. A pilot without an agreed failure point doesn’t fail. It lingers, using up the team’s credibility along with the budget.
Five questions before any pilot starts
Ask these before you sign. A “no” to any of the first four means not yet.
Real research doesn’t fit this. If you’re exploring what’s possible with no near-term number, say so and cap it as a research budget. These questions also can’t rescue a vendor that isn’t really selling an agent. I cover that check in what agentic AI is.
What to do on Monday
List your agent pilots. Write each one’s number, owner, baseline and kill date on a single page. Any pilot missing one gets 30 days to earn it or goes. It takes an afternoon.
At revenue-operations scale, also move every agent contract you can onto outcome-based or short terms before the renewals arrive. And pick one workflow, not one tool, for a real redesign in the second half of the year.
By the fourth quarter, every agent dollar should map to a number, an owner and a kill switch.
FAQ
What percentage of AI agent pilots reach production? McKinsey’s State of AI survey shows 62 percent of organisations experimenting with AI agents. 23 percent report scaling an agentic system anywhere in the enterprise. Fewer than 10 percent are scaling agents within any single function. Separately, Gartner predicts over 40 percent of agentic AI projects will be cancelled by the end of 2027. Roughly one in four experimenters reaching scale is a fair planning number.
Why do most AI agent projects fail? The main causes are organisational: no owned business metric, no baseline, no workflow redesign, and costs that grow past the pilot budget. Gartner cites escalating costs, unclear business value and weak risk controls as the drivers behind its 40 percent cancellation prediction. McKinsey’s high performers are 2.8x more likely to have fundamentally redesigned workflows around AI.
How do I measure ROI on an AI agent? Pick one target metric and its owner before the pilot starts. Measure the baseline first. Set a kill threshold with a date. Count full production costs (integration, evaluation, human review and monitoring as well as tokens) against how much the metric moves. If you can’t name the metric, the baseline and the kill criterion on one page, the pilot isn’t measurable and shouldn’t start.
Three-minute lesson on this Workflow or agent? Marketing course, advanced. Free, no sign-up.