Where the figures on AI pilots stuck in purgatory come from
The figures for AI pilots that never reach production come from four different measures: IDC's 88 percent, MIT's 95, S&P Global's 42 and Gartner's forecast of 30. Each counts something else. Read on 2026-11-06.
Contents
"Pilot purgatory" is the name given to AI pilots that run but never reach production. Articles quote the figure for it as 88 percent, 95 percent, 85 percent or 70 percent, as if one number were being repeated. The sources behind the figures measure different things on different samples: proofs of concept that did not reach deployment, pilots with no profit impact, companies abandoning most of their projects, and a forecast of projects abandoned after a proof of concept. This page lists each, with who produced it and what it counts. The IDC and S&P figures are read through press coverage, and the page says so where it matters.
IDC with Lenovo counts 88 percent of proofs of concept
CIO magazine, reporting on research by the analyst firm IDC "undertaken in partnership with Lenovo", wrote: "88% of observed POCs don't make the cut to widescale deployment. For every 33 AI POCs a company launched, only four graduated to production, IDC found."[1] The 88 percent is the same ratio: 29 of 33 did not graduate (this page's own arithmetic). The Lenovo-hosted IDC CIO Playbook 2025 is based on a survey of 2,920 IT and business decision makers and says: "The high number of AI POCs but low conversion to production indicates the low level of organizational readiness in terms of data, processes and IT infrastructure."[6] The playbook text, read in full, does not contain the 33-to-4 count. An IDC vice president told CIO: "These POCs are highly underfunded or not funded at all. Most of the time the POC happens not because of a strong business case."[1] What counts as "production" or "widescale deployment" is not defined in either source.
S&P Global Market Intelligence counts 42 percent abandoning most projects
CIO Dive, on an S&P Global Market Intelligence survey "of more than 1,000 respondents in North America and Europe", reported: "The share of companies abandoning most of their AI initiatives jumped to 42%, up from 17% last year. The average organization scrapped 46% of AI proof-of-concepts before they reached production."[2] This counts companies and their abandoned projects, not pilots stuck in limbo. S&P's own page could not be opened, so these figures rest on the press report alone.
MIT's NANDA initiative counts 95 percent
The report "The GenAI Divide: State of AI in Business 2025" states in its executive summary: "95% of organizations are getting zero return", and in the next sentences "Just 5% of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact."[4] A later section says "only 5% of custom enterprise AI tools reach production."[4] The first counts organizations, the second counts pilots, the third counts tools. The report's methodology: "52 structured interviews across enterprise stakeholders, systematic analysis of 300+ public AI initiatives and announcements, and surveys with 153 leaders. Success defined as deployment beyond pilot phase with measurable KPIs. ROI impact measured 6 months post-pilot".[4] It qualifies its pilot-to-production figures: "These figures are directionally accurate based on individual interviews rather than official company reporting."[4] Fortune's coverage describes the evidence differently: "150 interviews with leaders, a survey of 350 employees, and an analysis of 300 public AI deployments", which does not match the report's 52 interviews and 153 surveyed leaders.[3]
On buying versus building, the report says "external partnerships with learning-capable, customized tools reached deployment ~67% of the time, compared to ~33% for internally built tools", and adds that the figures "reflect self-reported outcomes and may not account for all confounding variables."[4] It also warns that the gap "may reflect organizational capabilities rather than implementation approach alone."[4]
Gartner gives a forecast of 30 percent
Gartner's press release of 29 July 2024 predicts: "At least 30% of generative AI (GenAI) projects will be abandoned after proof of concept by the end of 2025, due to poor data quality, inadequate risk controls, escalating costs or unclear business value."[5] It is a forecast made before the year ended, not a count.
How the four fit together
Synthesis, this page's own reading and not any source's claim. The four figures ask different questions: IDC asks what share of proofs of concept reached deployment; S&P asks what share of companies dropped most projects; MIT asks what share of pilots or companies saw profit impact; Gartner forecasts what share would be abandoned. A pilot can be alive and still earn nothing, so none of the four equals "stuck". The causes the sources name overlap: data quality or readiness, unclear business value, cost, and risk controls.[1][5]
The figures for agentic AI in particular, from Gartner, McKinsey and a vendor-run survey, are set side by side on Agentic AI: ROI figures and what they count. Three published frameworks for choosing which process to automate in the first place are on Automation opportunity assessment: three frameworks.
Not yet proven
- That these rates apply to small firms. The samples are enterprises and public deployments; none is a sample of small businesses.
- That the IDC 88 percent is an IDC-published figure. CIO attributes it to IDC; the IDC and Lenovo playbook text does not contain the 33-to-4 count.
- That the S&P figures are as reported. They rest on one press report; S&P's page could not be opened.
- That MIT's 95 percent is a measured rate. The report calls its pilot-to-production figures directionally accurate and based on interviews, and its own headline counts organizations, pilots and tools in different sentences.
- That buying beats building. The report calls the 67 and 33 percent self-reported and possibly due to organizational capability.
If you have run an AI pilot and can say what moved it into production or stopped it, I would like to compare notes: mackenzie@mackenziebowes.com.
Sources
- CIO, 88% of AI pilots fail to reach production, but that's not all on IT, 2025-03-25 (reporting IDC research with Lenovo). www.
cio. com/ article/ 3850763/ 88- of- ai- pilots- fail- to- reach- production- but- thats- not- all- on- it. html - CIO Dive, AI project failure rates are on the rise: report, 2025-03-14 (reporting S&P Global Market Intelligence). www.
ciodive. com/ news/ AI- project- fail- data- SPGlobal/ 742590/ - Fortune, MIT report: 95% of generative AI pilots at companies are failing, 2025-08-18 (reporting MIT NANDA). fortune.
com/ 2025/ 08/ 18/ mit- report- 95- percent- generative- ai- pilots- at- companies- failing- cfo/ - MIT NANDA (Aditya Challapally and colleagues), The GenAI Divide: State of AI in Business 2025, v0.1 PDF, file dated 2025-07-13; full text read. mlq.
ai/ media/ quarterly_decks/ v0. 1_State_of_AI_in_Business_2025_Report. pdf - Gartner, Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025, press release, 2024-07-29. www.
gartner. com/ en/ newsroom/ press- releases/ 2024- 07- 29- gartner- predicts- 30- percent- of- generative- ai- projects- will- be- abandoned- after- proof- of- concept- by- end- of- 2025 - Lenovo with IDC, CIO Playbook 2025: It's Time for AI-nomics, PDF, 36 pages, full text read after OCR. investor.
lenovo. com/ en/ global/ Lenovo_CIO_Playbook_2025. pdf
Terms on this page
- Agentic AI: AI that works toward a goal by planning steps, making decisions and acting with more or less independence.