The MIT report behind "95% of AI pilots fail," and what it measured

August 2025 — A report from an MIT Media Lab project says almost no companies are getting a return on generative AI. Here's what launched, what it cost, and what went wrong.

Cost and outcome (as of Aug. 26, 2025)

Spending: "$30–40 billion in enterprise investment into GenAI," the report says. (MIT NANDA report)
Headline finding: "95% of organizations are getting zero return." (MIT NANDA report)
Narrower figure: only 5% of custom enterprise AI tools reach production. (MIT NANDA report)
Caveat: the figures are "directionally accurate based on individual interviews rather than official company reporting." (MIT NANDA report)

What launched

The report, "The GenAI Divide: State of AI in Business 2025," comes from Project NANDA at MIT. Its authors are Aditya Challapally, Chris Pease, Ramesh Raskar and Pradyumna Chari, and its research ran from January to June 2025.

Its executive summary is blunt: "Despite $30–40 billion in enterprise investment into GenAI, this report uncovers a surprising result in that 95% of organizations are getting zero return." It adds: "Just 5% of integrated AI pilots are extracting millions in value."

Fortune's story on Aug. 18 put it this way: about 5% of AI pilot programs achieve rapid revenue acceleration, while most deliver little to no measurable impact on profit and loss.

What the report measured

The methodology section describes a review of over 300 publicly disclosed AI initiatives, interviews with representatives from 52 organizations and survey responses from 153 senior leaders, collected at four industry conferences. Fortune described the sample differently, as 150 interviews, a survey of 350 employees and 300 public deployments.

"Success" has a specific meaning. The report defines successfully implemented tools as ones users or executives said caused "a marked and sustained productivity and/or P&L impact." By that test, only 5% of custom enterprise AI tools reached production.

The report is open about its limits. Its figures, it says, are "directionally accurate based on individual interviews rather than official company reporting," and success definitions may differ across organizations. It warns of possible selection bias among companies willing to take part.

What else it found

Challapally, the lead author, told Fortune the problem was not model quality but a "learning gap" in how tools fit into company workflows. More than half of generative AI budgets go to sales and marketing tools, Fortune reported, yet the biggest return was in back-office automation.

Buying from specialized vendors succeeded about 67% of the time, while internal builds succeeded only one-third as often, according to Fortune. The report itself says this correlation "does not necessarily prove causation." Rather than mass layoffs, Fortune reported, companies were not backfilling roles as they became vacant.

The critics

Futuriom, a tech research firm that sells its own enterprise AI research, called the study "weird." It noted that NANDA, short for "Networked Agents and Decentralized AI," is a spinoff project from the MIT Media Lab that builds on Anthropic's and Google's agent protocols, and that some authors work at vendors.

Futuriom quoted Wharton professor Kevin Werbach, who wrote on LinkedIn that "there appears to be no further support for the 95% claim" and that he still couldn't understand where it comes from. Werbach's conclusion: "If MIT Project NANDA stands behind the claims, it should release the full supporting data. If not, it should retract the report."

What went wrong

A headline number without its caveats. Fortune's headline said 95% of pilots "are failing"; the report's own limits section calls its figures directional.
What we don't know yet: the full interview and survey data, how "zero return" was measured, and whether MIT will publish the underlying numbers.

Our follow-up question for MIT's Project NANDA

Will the authors release the interview and survey data behind the 95% figure?

Sources

Aditya Challapally, Chris Pease, Ramesh Raskar and Pradyumna Chari, MIT Project NANDA, “The GenAI Divide: State of AI in Business 2025,” July 2025 (copy hosted by mlq.ai). mlq.ai
Fortune, Sheryl Estrada, “MIT report: 95% of generative AI pilots at companies are failing,” Aug. 18, 2025. fortune.com
Futuriom, R. Scott Raynovich, “Why We Don't Believe MIT NANDA's Weird AI Study,” Aug. 26, 2025. futuriom.com

Disclosure: [not in copy] Editor placeholder