Key Findings

  • Nearly seven in ten (68%) U.S. businesses report that at least some of their artificial intelligenceprojects exceeded budgetover the past year, with one-third (33%) saying overruns occurred often or always. This data comes from a report released Wednesday by AI safety company WitnessAI.
  • The survey of 300 corporate executives also found that companies face difficulties in proving the return on investment of AI. Only 9% of respondents said that more than three-quarters of their AI projects delivered measurable financial returns.
  • The report states: "AI projects are exceeding budgets, falling short of ROI expectations, and hitting governance bottlenecks—bottlenecks that slow deployment without reducing risk."

Deeper Analysis

These findings highlight the growing challenge companies face in predicting and managing AI costs. With unforeseen expenses, limited visibility into spending, and rapidly evolving innovation and pricing models, it is increasingly difficult for businesses to control the financial performance of AI projects.

This challenge is becoming more pronounced as many organizations begin deploying AI at scale.

According to a report released by KPMG in June, the proportion of organizations orchestrating multiple AI agents across workflows doubled from 9% to 18% in the second quarter. Meanwhile, some companies are experimenting with what KPMG calls "unconventional and often counterproductive"approaches to driving AI usage。

The KPMG report shows that 41% of respondents said they would consider adopting a practice called "token-maxxing," which gamifies token consumption through incentives and leaderboards. 22% opposed the practice, while another 37% were neutral.

WitnessAI notes that, at the same time, the rise of "shadow AI"—where employees use AI applications outside formal IT and procurement processes—has also exacerbated cost overruns.

IT and infrastructure departments are the largest single source of shadow AI activity, accounting for 47%, higher than sales and business development (34%) and marketing and communications (33%).

WitnessAI states in the report: "The survey results show that the departments most responsible for governing AI usage within an organization are also the most likely to operate outside their own policies."

30% of respondents reported that unmanaged or poorly governed AI usage has led to cost overruns; 27% said it caused AI projects to be delayed or canceled.

WitnessAI says: "These findings reveal a gap in financial risk management: organizations are deploying AI agents faster than they are building the financial visibility, ownership structures, and risk management practices needed for responsible oversight."

At the same time, financial leaders are under increasing pressure to demonstrate that AI investments are delivering measurable returns.

A survey by CloudZero found that 87% offinancial leaders feel pressureto link AI spending to business outcomes within a year, but only 22% have already achieved this goal.

WitnessAI believes that part of the difficulty in demonstrating ROI from AI adoption is that it is rarely reflected in a single metric. The report says its impact "instead appears in fragmented form across budgets, productivity reports, and various business units, making a single ROI figure difficult to determine."