---
id: "claim-95-percent-failure"
type: "claim"
source_timestamps: ["¶2"]
tags: ["statistics", "roi"]
related: ["concept-experimentation-trap", "entity-mit"]
confidence: "high"
testable: true
speakers: ["Rens van den Broek", "Samantha Hellauer", "Dina Wang"]
source_url: "https://hbr.org/2025/09/what-companies-with-successful-ai-pilots-do-differently"
source_title: "What Companies with Successful AI Pilots Do Differently"
sources: ["execution"]
sourceVaultSlug: "hbr-seg-execution"
originDay: 8
articleStem: "hbr-foci-60-successful-ai-pilots"
sourceUrl: "https://hbr.org/2025/09/what-companies-with-successful-ai-pilots-do-differently"
sourceTitle: "What Companies with Successful AI Pilots Do Differently"
---
# 95% of Gen AI programs fail to deliver returns

## Claim: 95% of Gen AI programs fail to deliver returns

According to a recent [[entity-mit-d60|MIT]] report cited by the authors, a remarkable **95% of generative AI programs fail to deliver bottom-line returns**, highlighting a massive gap between technological experimentation and enterprise value creation. Only ~5% generate quantifiable value.

- **Confidence:** high
- **Testable:** yes

### Relationship
This statistic is the empirical motivation for the entire article and is the destination of the [[concept-experimentation-trap]].

### Enrichment — external validation
Multiple secondary summaries of the **MIT Project NANDA / Media Lab 'AI in Business 2025'** report confirm the ~95% figure and the 'bottom-line returns' framing (Fortune, Forbes, CloudFactory). Practitioner write-ups specify the study examined **300+ initiatives** and defines failure as *no documented, defensible business value*, not merely low ROI.

### Enrichment — nuance / counter-perspective
The methodology has been questioned (notably by Ethan Mollick): it rests on ~52 interviews with convenience sampling and a **six-month impact window**, which some argue is too short for complex enterprise transformations. Treat 95% as a **strong directional signal, not a universal law**.


## Related across articles
- [[claim-marginal-business-impact]]
- [[claim-widening-performance-gap]]
- [[question-laggard-catchup-viability]]
- [[contrarian-ai-hype-vs-reality]]