The short answer
AI pilots stall when they start with a tool rather than a question. Only about 6 percent of organizations attribute 5 percent or more of their EBIT to AI7, and Gartner expects more than 40 percent of agentic AI projects to be canceled by the end of 20278. Pilots that pay start with a written business assumption, a number on it, and a stop condition, and use AI only to answer that question.
The adoption gap is real
Larger firms use AI at roughly twice the rate of businesses overall: about 37 percent of US firms with 250 or more employees, against 17 to 20 percent of all firms, according to the Census Bureau’s Business Trends and Outlook Survey6. Using AI and earning from it are different things, though. McKinsey’s 2026 survey found only about 6 percent of organizations attribute 5 percent or more of EBIT to AI7.
Three reasons pilots stall
- They start with a tool. A team buys or builds something capable, then searches for a problem it can solve. The problem they find is usually the one easiest to demo, not the one worth the most.
- Nobody sized the prize first. Without a dollar range agreed in advance, a pilot can only be judged on whether it works, not on whether it was worth doing.
- There is no way to stop. Pilots without a written end condition drift. They keep a small team busy and never quite become a decision.
Start with the question instead
A written assumption
Your records
A sized answer
A tool, if one is needed
The order matters more than the technology. Fewer than 30 percent of digital transformations succeed12, and people weigh a loss about twice as heavily as an equal gain11, so a buyer’s real question is rarely what AI could do. It is what this particular project could cost if it is wrong. Starting with the question answers that before any money is committed.
A useful test
Before approving any AI pilot, ask one thing: what number will this move, by how much, and what result would make us stop? If nobody can answer, the pilot is not ready.
Sources
- 6US Census Bureau, Business Trends and Outlook Survey, reported May 26, 2026 (Dec 14, 2025 – May 3, 2026).
- 7McKinsey & Company, The State of AI, August 2026.
- 8Gartner, press release, June 25, 2025.
- 11Tversky, A. & Kahneman, D. (1992). Advances in prospect theory. Journal of Risk and Uncertainty, 5, 297–323.
- 12McKinsey & Company, “Unlocking success in digital transformations,” October 2018.
Market figures describe the landscape and are not Modven results.