MIT: 95 percent of AI pilots show no measurable impact, Gartner expects many canceled projects

Many companies have jumped into AI, but far from all efforts pay off on the bottom line. Two analyses paint a fairly sober picture, though they measure different things: One is a tally of what has happened, the other is a forecast.

MIT: Few pilots deliver measurable impact

In summer 2025, MIT's NANDA initiative published the report "The GenAI Divide: State of AI in Business 2025." According to Fortune's account, it shows that 95 percent of companies' generative AI pilots deliver no measurable impact on the income statement. Generative AI is the type of AI that can write text and create images, such as ChatGPT. Only 5 percent of pilot programs achieve rapid revenue growth.

The figure does not mean that 95 percent of AI products on the market fail. It concerns companies' pilot projects, meaning smaller trials, and the fact that they have not left a clear mark on the financial statements.

The main explanation is a "learning gap," says lead author Aditya Challapally of MIT. Generic tools like ChatGPT do not adapt to companies' workflows. The authors place less emphasis on the quality of the models.

According to the report, more than half of AI budgets go to sales and marketing, even though the return is better from automating the back office, meaning the internal support functions. Projects using purchased, specialized AI solutions succeeded in about 67 percent of cases, compared with about 33 percent for internal builds.

Some large companies and younger startups do very well, however. According to Challapally, startups have seen revenue climb from zero to $20 million in a year.

Gartner: Cancellations ahead

In a press release dated June 25, 2025, Gartner predicts that more than 40 percent of agentic AI projects will be canceled by the end of 2027. Agentic AI refers to systems that are meant to carry out tasks on the user's behalf. According to Gartner, the causes are rising costs, unclear business value and inadequate risk controls.

Anushre Verma, senior director analyst at Gartner, says most agentic projects today are early experiments or proofs of concept, driven mostly by hype and often misapplied. Gartner also believes many vendors engage in "agent washing": They rebrand chatbots and RPA without real agentic capabilities. Only about 130 of the many thousands of vendors are judged to be genuine.

In a Gartner poll from January 2025 of 3,412 webinar attendees, 19 percent said their organization had made significant investments in agentic AI. 42 percent had invested cautiously, and 8 percent had not invested at all. 31 percent were waiting or unsure.

Gartner also expects 15 percent of day-to-day work decisions to be made autonomously in 2028 (0 percent in 2024), and 33 percent of enterprise software to include agentic AI in 2028 (under 1 percent in 2024).

Caveats

The figures come from intermediaries and not from the original documents. The MIT report's sample is relatively small. According to Fortune, it is based on 150 interviews, a survey of 350 employees and 300 publicly known AI deployments, while another account mentions 52 interviews and 153 survey responses. Fortune also writes that companies were reluctant to share failure rates. Gartner's figures are a forecast, and Gartner itself sells advisory services.