AI project failure rates are on the rise (CIO Dive on S&P Global, March 2025)
TL;DR
A short CIO Dive news brief by Lindsey Wilkinson, 14 March 2025. It reports a survey by S&P Global Market Intelligence. The wiki has not seen the survey itself: S&P Global’s own pages refused access. Every figure below is S&P’s as CIO Dive reports it.
- 42% of companies abandoned most of their AI initiatives, “up from 17% last year.”
- The average organisation scrapped 46% of AI proofs of concept before they reached production.
- The top obstacles were cost, data privacy and security risks.
- Use is concentrated in IT operations, then customer experience and marketing.
- The sample is “more than 1,000 respondents in North America and Europe.”
Key data
| Measure | Figure | As stated by CIO Dive |
|---|---|---|
| Companies abandoning most AI initiatives | 42% (previous survey: 17%) | “jumped to 42%, up from 17% last year” |
| Proofs of concept scrapped before production | 46% (average organisation) | “scrapped 46% of AI proof-of-concepts” |
| Sample | >1,000 respondents | North America and Europe |
| Top obstacles | Cost, data privacy, security risks | |
| Most common uses | IT operations, customer experience, marketing |
Method and definitions
The brief gives no survey name, fieldwork dates, respondent roles or question wording. It does not say what counts as an “initiative” or as “abandoning most” of them, or whether “AI” means generative AI only. It does not link to the source. Search-engine summaries name the survey as 451 Research’s Voice of the Enterprise: AI & Machine Learning, Use Cases 2025 and give a sample of 1,006 fielded in October-November 2024. Those details could not be checked and are not used here.
The rest of the brief is context: an Informatica report that two-thirds of enterprises cannot move generative-AI pilots into production, and two quotes from a CIO Dive virtual event. Amanda Luther of Boston Consulting Group: “Celebrating failures on some of this matters.” Htike Htike Kyaw Soe of KPMG U.S.: “When people hear the word failure, they freak out about it.”
Scope and reliability
- Secondary. A news brief about a vendor-research survey, read without the survey. The 42% and 46% are as reported, not verified.
- Year labels. CIO Dive writes “this year” (2025) and “last year” (2024). If the unverified fieldwork dates are right, the 42% describes firms surveyed in late 2024 and published in 2025. Quote it as “the 2025 survey”, not as “in 2025, 42% stopped.”
- Undefined measure. “Abandoning most AI initiatives” is a firm-level share; “46% of proofs of concept” is a project-level average. They are different measures.
- Commercial research, sold by S&P Global Market Intelligence. Self-reported by IT and business respondents.
Dynamic-capabilities reading
digital-seizing/balancing-digital-portfolios: the figures describe how few AI projects firms scale up: nearly half of proofs of concept stop before production, and 42% of firms dropped most of their initiatives.contextual/internal-barriers: cost, data privacy and security risks are named as the main obstacles.
How it connects
- enterprise-ai-adoption and the thread ai-maturity-measurement-comparison: a pilot-to-production failure figure, set against the much higher adoption figures in executive surveys.
- BBC’s AI Decoded quotes a vendor index with about 30% of pilots in production, and the MIT 95% figure.
- Allen (AWS) names the same pattern “pilot purgatory”.
- micro-productivity-trap: adoption without value, here as projects stopped before production.
- MIT NANDA, The GenAI Divide: a 2025 report on the same drop-off from pilot to production, with its own, stricter definition of success.
What was actually ingested
The full CIO Dive brief: deck, Dive Brief and Dive Insight. Two S&P Global pages that appear to carry the survey (a “Highlights from VotE: AI & Machine Learning” page and an October 2025 research page) and a Seeking Alpha repost all returned HTTP 403 and were not read.
Linked entities and concepts
- Entities: Boston Consulting Group (Amanda Luther, quoted).
- Concepts: enterprise-ai-adoption, micro-productivity-trap.
- Dangling (single-source mention, deferred): Lindsey Wilkinson, CIO Dive, S&P Global Market Intelligence, 451 Research, Amanda Luther, Htike Htike Kyaw Soe, KPMG U.S., Informatica.