Key Findings

  • Businesses generally struggle to predict AI costs, with spending growing faster than most organizations can track, according to a survey commissioned by Mavvrik, an AI cost management company.
  • The risks could escalate sharply as many companies scale up AI investments next year, said Sundeep Goel, CEO of Mavvrik.
  • "I personally think Q1 and Q2 of 2026 will be when the problem really shows up, when a lot of pilots move into production," Goel said in an interview. "My prediction is that CFOs will face significant cost overruns."

Detailed Analysis

The survey data shows that more than half (56%) of companieshave AI cost forecast errors between 11% and 25%, while nearly a quarter (24%) have errors exceeding 50%. 84% of respondents said AI costs caused gross margin losses of more than 6%, with over a quarter of companies experiencing gross margin losses of 16% or more.

"These numbers should be a wake-up call for every finance leader," said Ray Rike, CEO of Benchmarkit, a software-as-a-service metrics company, in a press release. "AI is no longer just experimental technology—it is hitting gross margins, and most companies can't even predict its impact."

According to Ernst & Young, corporate AI budgets have risen sharply over the past year as agentic technology has injected new momentum into the market.

A study released by EY in July shows that about 21% of surveyed companiesare investing $10 million or more in AI, up from 16% a year earlier; another third (35%) expect to spend $10 million or more on the technology next year.

Consulting firm Catalysis Group noted in a March article that organizations that push forward with AI without fully understanding hidden costs mayface unexpected financial pressure。

Catalysis said that in many areas, companies tend to underestimate costs related to data acquisition, storage, cleaning, and security.

"An engagement program that starts out fun often reveals significant data challenges rooted in historical business practices and systems," the article stated.

Mavvrik's research also found no significant difference between large and small companies in the likelihood of major forecast misses.

"Companies with revenue between $10 million and $50 million are most likely to keep errors within ±10%, possibly because they are AI-native enterprises that established granular cost tracking early on," Mavvrik said in the report. The report is based on a survey of 372 organizations across industries and revenue tiers.