AI Project Failure Statistics for 2026: Every Major Number, Sourced and Sorted

There is no single AI project failure rate, and the most useful figures are the ones that say exactly what they counted. The strongest recent measurements put outright failure and abandonment in a fairly narrow band. Gartner's survey of 782 infrastructure and operations leaders, published in April 2026, found that only 28% of their AI use cases fully succeed and meet ROI expectations, while 20% fail outright. S&P Global found that 42% of companies abandoned most of their AI initiatives before production in 2025, up from 17% a year earlier, and that the average company scrapped 46% of its proofs of concept along the way.

The famous numbers sit well above that band, at 80%, 85%, 87%, 88%, and 95%, and each one counts something narrower than its reputation suggests, comes to us secondhand, or has no study behind it at all. Sorted by what they actually measure (organizations, projects, pilots, or predictions), the figures stop contradicting each other and start describing one consistent problem. If you want the reasons behind the numbers rather than the numbers themselves, our pillar guide on why AI projects fail covers the five causes that keep coming up.

Key Takeaways

• No study measures a universal AI project failure rate. Published figures range from about 20% to 95% because they count different units: organizations, projects, pilots, or forecasts.
• The best recent measurements: Gartner (April 2026) found 28% of AI use cases in IT operations fully succeed and 20% fail outright; S&P Global (May 2025) found 42% of companies abandoned most AI initiatives, up from 17%, and companies scrapped 46% of proofs of concept on average.
• Getting pilots into production is the main choke point. Gartner's 2024 survey found 48% of AI projects reach production on average, taking eight months, and IDC found only 4 of every 33 proofs of concept made it.
• Return is a separate problem from abandonment. MIT's 95% counts organizations with no measurable P&L return from generative AI, and McKinsey's 2026 survey found only 37% of respondents attribute any EBIT impact to AI, unchanged from a year earlier.
• The "87% of data science projects never reach production" figure comes from a 2019 conference panel write-up that cites no study, and Gartner's "85%" was a 2018 forecast about biased outputs.
• No government survey measures small businesses that tried AI and stopped. The Federal Reserve found 46% of small employer firms use AI, but only 7% of those users have fully integrated it.

Table of Contents

• How many AI projects fail? The short answer
• Every major AI failure statistic, sorted by what it counts
• Abandonment and pilot-to-production statistics
• AI return-on-investment statistics
• Analyst predictions about AI project failure
• Statistics on why AI projects fail
• AI agent failure statistics
• Are AI projects failing more often than before?
• Small business AI statistics
• How AI failure rates compare with ordinary IT projects
• Famous figures and where they really come from
• Five questions to ask of any AI failure statistic
• What the numbers mean for a 10-to-200-person business
• Where Avolis fits
• Frequently Asked Questions
• Continue Learning

How Many AI Projects Fail? The Short Answer

Most measured rates put outright failure or abandonment at somewhere between a fifth and about half of AI projects, depending on the survey and its definition, and a larger share never produce a return anyone can measure. The one measured figure far outside that band, IDC's 88% of proofs of concept, comes from unpublished research known only through press coverage. That two-part answer matters because the most quoted statistics blur abandonment and missing returns together, and a business owner deciding whether to start a project needs to know which risk they're actually facing.

The abandonment side has three solid measurements. Gartner's 2026 survey found 20% of AI use cases in IT operations fail outright (Gartner, "Gartner Says AI Projects in I&O Stall Ahead of Meaningful ROI Returns", April 2026). S&P Global's 2025 survey of 1,006 IT and business professionals found the average company scrapped 46% of its AI proofs of concept before broad adoption (S&P Global Market Intelligence, Voice of the Enterprise: AI & Machine Learning, May 2025). Gartner's 2024 survey of 644 respondents found that only 48% of AI projects make it into production on average (Gartner, "Gartner Survey Finds Generative AI Is Now the Most Frequently Deployed AI Solution in Organizations", May 2024).

The return side looks worse, because a project can survive to production and still change nothing on the income statement. MIT's 95% and McKinsey's finding that only 37% of respondents attribute any EBIT impact to AI both sit on this side, and both describe organizations rather than individual projects.

Every Major AI Failure Statistic, Sorted by What It Counts

The 23 figures below are the failure and success rates most often cited in AI planning conversations, grouped by the unit each one counts. A figure about organizations can't be compared directly with a figure about projects, and a forecast carries less weight than a survey result, whatever the size of the number.

Across the groups, the measured project-level figures for outright failure or non-deployment mostly fall between 20% and about half, with IDC's 88% as the outlier. The organization-level figures about returns run much higher, because a company with ten AI projects can report no measurable return even when two of them are working fine.

Abandonment and Pilot-to-Production Statistics

Most AI work that fails dies between the pilot and the production system, and the best data on that stage comes from surveys of large organizations. In S&P Global's 2025 data, a typical company dropped close to half (46%) of its proofs of concept before broad adoption, and 42% of companies walked away from the majority of their AI work before it went live. Firms with higher failure rates reported more pushback from employees and customers than the rest.

Gartner's 2024 survey of 644 respondents in the US, Germany, and the UK found that on average 48% of AI projects make it into production, and that the trip from prototype to production takes eight months. IDC's research with Lenovo, as reported by CIO in March 2025, put the rate lower, with only 4 of every 33 AI proofs of concept reaching wide deployment, which is where the widely shared 88% comes from. IDC's authors linked the low conversion to weak organizational readiness in data, processes, and IT infrastructure.

Deloitte's January 2026 survey of 3,235 business and IT leaders adds a timing detail that explains a lot of the gap. Only 25% of respondents had moved 40% or more of their AI pilots into production, while 54% expected to reach that level within the next three to six months (Deloitte, State of AI in the Enterprise, January 2026). More than twice as many leaders expected to reach that level within months as had reached it already.

MIT's funnel tells the same story at the organization level, at 60% investigating a task-specific generative AI tool, 20% piloting, and 5% implementing successfully, which works out to roughly one successful pilot in four (Challapally, Pease, Raskar, and Chari, The GenAI Divide: State of AI in Business 2025, MIT NANDA, July 2025).

AI Return-on-Investment Statistics

Getting an AI tool into production doesn't guarantee it pays for itself, and the return statistics are where the scariest headline numbers live. MIT NANDA's 95% is the best known, and it describes organizations that couldn't point to measurable P&L impact from generative AI. MIT's bar for success was impact that users or executives would call "marked and sustained," and the report labels its own funnel numbers directionally accurate. The pillar guide walks through how that definition shapes the headline.

In McKinsey's 2026 survey of 1,719 participants, roughly 80% reported personal productivity gains from AI, but only 37% could tie any EBIT impact to it, about the same as in 2025 (McKinsey, The State of AI in 2026: On the Road to ROI, August 2026).

BCG's 1,250-executive survey sorted companies into three tiers: a 60% majority it called laggards, with minimal revenue and cost gains to show, a 35% middle group starting to see value, and a 5% top tier BCG called "future-built" (BCG, "AI Leaders Outpace Laggards with Double the Revenue Growth and 40% More Cost Savings", September 2025). IBM's 2025 study of 2,000 CEOs found the CEOs themselves estimated that only 25% of their AI initiatives had delivered the expected return and only 16% had scaled across the enterprise (IBM Institute for Business Value, "IBM Study: CEOs Double Down on AI While Navigating Enterprise Hurdles", May 2025).

The newest figures come from Gartner's HR practice, released on 26 October 2026. Gartner said 97% of organizations are unable to achieve the cost savings they want through productivity alone, and analyst CV Viverito said in the keynote that 38% of companies are over budget on AI while 85% plan to spend more next year. A Gartner survey of 1,865 employees in the second quarter of 2026 found that only 21% of the time AI freed up was redirected into more innovative work (Gartner, "Gartner HR Research Finds 77% of Board Members Expect AI to Deliver Productivity and Cost Savings, Yet Just 3% of Organizations Exceed Cost Savings Goals Via Productivity", October 2026). When the hours a tool saves aren't assigned to other work, the savings don't reach the income statement, and a tool that works can still be counted as returning nothing.

Analyst Predictions About AI Project Failure

Many of the most shared AI failure statistics are forecasts, and two of the best known have already reached their deadlines. A forecast tells you what an analyst firm expected to happen, so it belongs in a planning conversation, but it shouldn't be quoted as a rate that was observed.

Gartner predicted in July 2024 that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs, and unclear business value (Gartner, "Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025", July 2024). Its February 2025 forecast, which runs through the end of 2026, is that 60% of AI projects lacking AI-ready data will be abandoned. The survey behind it covered 1,203 data management leaders (Gartner, "Lack of AI-Ready Data Puts AI Projects at Risk", February 2025).

On AI agents, meaning tools that carry out steps in other systems instead of only drafting text, Gartner predicted in June 2025 that over 40% of agentic AI projects will be canceled by the end of 2027 because of escalating costs, unclear business value, or inadequate risk controls (Gartner, "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027", June 2025). In September 2026 it went further, predicting that by 2028, 70% of enterprises will abandon agentic AI built by vendor engineers embedded with them, "trapped by soaring costs and unable to evolve it on their own" (Gartner, "Gartner Predicts 70% of Enterprises Will Abandon Agentic AI Built by Vendor Forward-Deployed Engineering by 2028", September 2026). Its advice in that release was to set knowledge transfer, ownership, and an exit plan from the first day of an engagement.

Statistics on Why AI Projects Fail

The cause statistics agree more than the rate statistics do, and they mostly point at decisions made before anything is built. RAND interviewed 65 data scientists and engineers in 2024, and 84% of them named a leadership-driven cause, such as a misunderstood problem, as the main reason projects fail (Ryseff, De Bruhl, and Newberry, The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed, RAND, 2024).

Gartner's 2026 survey of I&O leaders puts proportions on the same pattern. Of those who reported at least one failure, 57% said they had expected too much, too fast. Skills gaps and poor or unavailable data were each cited by 38% of leaders who faced setbacks, while the 77% with at least one success credited it mainly to integrating AI into existing workflows and systems and to full support from business executives.

Two of those rows come from vendors. Salesforce sells integration software and Writer sells an enterprise AI platform, so their numbers describe enterprise conditions rather than project outcomes (Salesforce, "Salesforce Announces 2026 Connectivity Report", February 2026; Writer, "Key findings from our 2025 enterprise AI adoption report", March 2025). BARC's figure comes from its Lessons from the Leading Edge study of organizations delivering AI (BARC, "Global Study Maps the Reality of AI Delivery in 2025", December 2025). Our page on AI project data quality failure covers how to tell whether data really caused a specific failure.

AI Agent Failure Statistics

AI agents still fail most multi-step tasks in controlled tests, because one wrong step early in a task carries through every step after it. In Carnegie Mellon's TheAgentCompany benchmark, a mock software firm complete with internal sites, documents, and simulated colleagues, the strongest agent tested finished 30% of its assigned tasks without help (Xu et al., TheAgentCompany, arXiv:2412.14161, revised September 2025).

Salesforce AI Research's CRMArena-Pro benchmark tested leading agents on business tasks in a Salesforce environment with synthetic data. They succeeded on about 58% of tasks completed in a single exchange, and that fell to about 35% when the task took several turns of back-and-forth, much like a real customer or coworker conversation (Huang et al., CRMArena-Pro, arXiv:2505.18878, May 2025). Benchmarks aren't field results, but they're consistent with Gartner's view that many agentic projects will be cancelled. Our guide to why agentic AI projects fail covers the agent-specific causes.

Are AI Projects Failing More Often Than Before?

Yes, by the one measure repeated in consecutive years by the same survey. S&P Global's abandonment figure more than doubled between 2024 and 2025, from 17% to 42%, and reported positive impact slipped on each objective it tracks, by about five points on both revenue growth and cost management.

Other trend lines point the same way without measuring abandonment directly. McKinsey's EBIT figure didn't move between its 2025 and 2026 surveys, even as reported productivity gains spread. Stanford's 2026 AI Index reports that the AI Incident Database recorded 362 documented AI incidents in 2025, up from 233 in 2024 (Stanford HAI, 2026 AI Index Report, Responsible AI chapter, April 2026). Incidents are publicly reported harms rather than cancelled projects, but the count rises as more AI goes into live use.

Earlier in 2025, IDC's Ashish Nadkarni told CIO that many generative AI pilots were launched under board pressure with little or no funding. Our page on why AI projects are failing covers what changed in the 2025 and 2026 wave.

Small Business AI Statistics

Almost every AI failure statistic comes from large organizations, and none of the major government business surveys in the US or Canada tracks firms that tried AI and then dropped it. What those surveys do show is how shallow most small-business AI use still is.

In the Federal Reserve's 2025 Small Business Credit Survey of 6,525 firms with 1 to 499 employees, 46% reported some AI use. Only 7% of those users had built it fully into their processes, and 83% used it for writing or marketing (Federal Reserve Banks, 2026 Report on Employer Firms, March 2026). Accuracy (46%) topped the list of problems for current users, and fit with the business (54%) topped the list of worries for firms planning to start.

The Census Bureau's Business Trends and Outlook Survey found that between December 2025 and May 2026, 17% to 20% of US businesses used AI in some business function. Use reached 37% among firms with at least 250 employees and 32% among firms with 100 to 249, while firms with fewer than 20 employees showed no significant increase over the period (US Census Bureau, "Large Firms With at Least 20 Employees Biggest AI Users", May 2026).

The failure research has one size split, and it favors smaller firms. In MIT's data, the best mid-market companies went from pilot to full rollout in about 90 days, while enterprises with more than $100 million in revenue took nine months or more. Industry matters too: S&P Global reported the worst skills gaps, and lower returns, in manufacturing, construction, and real estate.

How AI Failure Rates Compare With Ordinary IT Projects

AI projects fail at higher rates than ordinary IT projects by the usual estimate, but the comparison is weaker than it's often made to sound. RAND's report says AI projects fail at twice the rate of IT projects without AI, and its footnote attributes that comparison to a 2023 Harvard Business Review article rather than to RAND's own research.

Ordinary IT projects set a low bar of their own. McKinsey and the University of Oxford studied more than 5,400 IT projects and found that large ones (initial budgets above $15 million) ran 45% over budget and 7% over schedule on average, while delivering 56% less value than predicted (Bloch, Blumberg, and Laartz, "Delivering large-scale IT projects on time, on budget, and on value," McKinsey, October 2012). Earlier research by Keil, Mann, and Rai found that 30% to 40% of information systems projects show some degree of escalation, meaning a troubled project keeps getting money and time after the evidence says it shouldn't (Keil, Mann, and Rai, "Why Software Projects Escalate," MIS Quarterly, 2000). Loose scope and runaway commitment sank IT projects long before AI, and the same fixes (a defined scope, a measured baseline, and a stop rule agreed in advance) apply. Our page on recovering a failing AI project covers how to apply them midstream.

Famous Figures and Where They Really Come From

The most repeated AI failure statistics each say something narrower in the original, and the 87% has no underlying study at all. The pillar guide covers the 85%, 80%, and 95% figures in depth.

The 87% has no study behind it: VentureBeat's July 2019 article, "Why do 87% of data science projects never make it into production?", reports on a panel at its own Transform conference with IBM's Deborah Leff and Gap's Chris Chapo (VentureBeat, July 2019). The article states that "only 13%" of data science projects make it into production, then describes that as "just one out of every 10," and it names no survey, sample, or method. The figure predates ChatGPT by three years and covers data science projects rather than AI tools bought from a vendor. It still circulates in 2026 vendor content as a current AI statistic.

Five Questions to Ask of Any AI Failure Statistic

Before you repeat an AI failure figure to a partner, a lender, or your own team, five questions will tell you what it can and can't support. Most misleading uses fail the first or second one.

1. What unit does it count? Organizations, projects, pilots, use cases, and tasks give very different numbers. An organization with one working tool and four abandoned pilots can count as a success, a failure, or 80% failed, depending on the unit.
2. What does "failure" mean here? Abandoned before production, never reached production, no measurable return, erroneous outputs, and over budget are five different things, and each of the famous figures uses one of them.
3. Was it measured, predicted, or borrowed? A survey result, an analyst forecast, and a figure cited from someone else's article carry different weight. Check whether a forecast's deadline has already passed.
4. Who was asked, and how big were they? Most samples are enterprises with large budgets and IT departments. A finding about $15 million programs tells a 40-person firm about direction, not odds.
5. When was it collected? A 2019 figure about data science projects says little about a 2026 tool that drafts quotes inside your job management system.

What the Numbers Mean for a 10-to-200-Person Business

Read for what they count, the statistics describe projects chosen and scoped badly far more than a technology too risky for a smaller operation. S&P Global's lower-failure organizations did more of their checking up front, looking at risk, compliance, and whether the data existed before a project was approved. Gartner's successful I&O leaders credited integration into existing workflows and executive support, which are both choices made before the build.

Consider the owner of a 60-person HVAC service company (an illustrative composite) whose business partner has read that "95% of AI fails." The partner's worry is answered best by a project scoped tightly enough to judge. That means naming the one workflow the project should change (say, the office coordinator's six hours a week of rebuilding technician notes into invoices), measuring that before anything is bought, deciding which system the output lands in, and agreeing in advance what result by what date would justify continuing. A project set up that way can still fail, but it fails within weeks and on a budget the owner chose, and the company learns something specific about its own invoicing.

If you want a structured way to do that selection work across several workflows at once, our guide to AI readiness assessment services covers what a good assessment includes, and our page on corporate AI implementation failure covers the organizational side of adoption.

Where Avolis Fits

Nearly every statistic above is missing the same thing at the company level: a before-number for the work the AI was meant to change. Every engagement starts with the diagnostic, which produces that number. Working with the staff who own each candidate workflow, we record how much of the work there is, how long each step takes, and where it changes hands, then score each workflow on payoff against effort. The scored list is yours to keep, and it stands on its own whether or not anything gets built. Where a workflow earns a build, we put it into the tools your staff already work in and stay on through the first months, measuring against the same before-number.

For the questions to put to any firm you're considering, see which AI consulting company to choose.

Frequently Asked Questions

What is the AI project failure rate in 2026?

No single rate exists. Gartner's 2026 survey found 20% of AI use cases in IT operations fail outright and 28% fully succeed. S&P Global found companies scrapped 46% of AI proofs of concept on average, and 42% of companies abandoned most initiatives in 2025, up from 17% the year before.

Is it true that 95% of AI projects fail?

Not as usually stated. MIT NANDA's 2025 report found 95% of organizations saw no measurable P&L return from generative AI, using a strict test of "marked and sustained" impact. It counts organizations, not projects, and among organizations that ran a pilot, about one in four implemented the tool successfully.

How many AI pilots make it to production?

Gartner's 2024 survey of 644 respondents found 48% of AI projects reach production on average, taking about eight months from prototype. IDC, as reported by CIO, found only 4 of every 33 proofs of concept reached wide deployment. Deloitte found just 25% of organizations had moved 40% or more of their pilots into production.

Where does the "87% of AI projects fail" statistic come from?

It comes from a July 2019 VentureBeat article about a panel at its Transform conference, which said only 13% of data science projects reach production. The article cites no survey or method, and the figure predates ChatGPT. Gartner's and IDC's measured production rates are better figures to quote.

Are AI projects failing more often than before?

The latest year-over-year measure, from 2024 to 2025, says yes. S&P Global found companies abandoning most AI initiatives rose from 17% in 2024 to 42% in 2025, and reported positive impact fell on every objective. McKinsey's 2026 survey found the share reporting any EBIT impact from AI flat at 37%.

Continue Learning

The AI project failure statistics worth repeating are the ones that state their unit, their definition, and their sample, and on those terms the picture is consistent. Most measured rates put abandonment or outright failure between a fifth and about half of projects, most of the losses happen between pilot and production, and a larger share of organizations can't yet point to a return.

Before you quote a failure rate in your next planning meeting, check which of the five questions it can answer.

Why projects fail, in detail:

• Why do AI projects fail? Common reasons and statistics
• AI project data quality failure
• Corporate AI implementation failure
• Why are AI projects failing?
• Why agentic AI projects fail

When a project is already in trouble:

• Why your AI consulting project isn't reaching production
• How to recover a failing AI project

Starting the right way:

• AI readiness assessment services
• Which AI consulting company should I choose?

---

Sources

• Gartner, "Gartner Says AI Projects in I&O Stall Ahead of Meaningful ROI Returns," press release, 7 April 2026 (survey of 782 infrastructure and operations leaders, November to December 2025), retrieved 2026-10-26: https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-says-artificial-intelligence-projects-in-infrastructure-and-operations-stall-ahead-of-meaningful-roi-returns
• Gartner, "Gartner HR Research Finds 77% of Board Members Expect AI to Deliver Productivity and Cost Savings, Yet Just 3% of Organizations Exceed Cost Savings Goals Via Productivity," press release, 26 October 2026 (includes a July 2026 survey of 297 CHROs and a 2Q26 survey of 1,865 employees), retrieved 2026-10-26: https://www.gartner.com/en/newsroom/press-releases/2026-10-26-gartner-hr-research-finds-77-percent-of-board-members-expect-ai-to-deliver-productivity-and-cost-savings-yet-just-3-percent-of-organizations-exceed-cost-savings-goals-via-productivity
• S&P Global Market Intelligence, "AI experiences rapid adoption, but with mixed outcomes: Highlights from VotE: AI & Machine Learning," 30 May 2025 (survey of 1,006 midlevel and senior IT and line-of-business professionals, North America and Europe), retrieved 2026-10-08: https://www.spglobal.com/market-intelligence/en/news-insights/research/ai-experiences-rapid-adoption-but-with-mixed-outcomes-highlights-from-vote-ai-machine-learning
• Gartner, "Gartner Survey Finds Generative AI Is Now the Most Frequently Deployed AI Solution in Organizations," press release, 7 May 2024 (survey of 644 respondents in the U.S., Germany, and the U.K., fielded Q4 2023), retrieved 2026-10-08: https://www.gartner.com/en/newsroom/press-releases/2024-05-07-gartner-survey-finds-generative-ai-is-now-the-most-frequently-deployed-ai-solution-in-organizations
• CIO, "88% of AI pilots fail to reach production — but that's not all on IT," 25 March 2025 (reporting IDC research conducted with Lenovo), retrieved 2026-10-08: https://www.cio.com/article/3850763/88-of-ai-pilots-fail-to-reach-production-but-thats-not-all-on-it.html
• Deloitte, "State of AI in the Enterprise" 2026 press release, January 2026 (survey of 3,235 business and IT leaders in 24 countries, fielded August to September 2025), retrieved 2026-10-08: https://www.deloitte.com/us/en/about/press-room/state-of-ai-report-2026.html
• Aditya Challapally, Chris Pease, Ramesh Raskar, and Pradyumna Chari, The GenAI Divide: State of AI in Business 2025, MIT NANDA, July 2025 (preliminary findings; 300+ public AI initiatives reviewed, interviews with 52 organizations, survey of 153 senior leaders; pages 3, 6, and 7 cited), retrieved 2026-10-08: https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf
• Dan Tinkoff, Lieven Van der Veken, and Michael Chui, with Tara Balakrishnan, The State of AI in 2026: On the Road to ROI, McKinsey & Company, 25 August 2026 (online survey of 1,719 participants in 97 nations, fielded 4 May to 8 June 2026), retrieved 2026-10-08: https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
• Boston Consulting Group, "AI Leaders Outpace Laggards with Double the Revenue Growth and 40% More Cost Savings," press release, 30 September 2025, for The Widening AI Value Gap: Build for the Future 2025 (survey of 1,250 senior executives and AI decision makers), retrieved 2026-10-08: https://www.bcg.com/press/30september2025-ai-leaders-outpace-laggards-revenue-growth-cost-savings
• IBM Institute for Business Value, "IBM Study: CEOs Double Down on AI While Navigating Enterprise Hurdles," press release, 6 May 2025 (2,000 CEOs from 33 countries and 24 industries, surveyed with Oxford Economics, February to April 2025), retrieved 2026-10-08: https://newsroom.ibm.com/2025-05-06-ibm-study-ceos-double-down-on-ai-while-navigating-enterprise-hurdles
• Gartner, "Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025," press release, 29 July 2024, retrieved 2026-10-08: https://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
• Gartner, "Lack of AI-Ready Data Puts AI Projects at Risk," press release, 26 February 2025 (survey of 1,203 data management leaders, July 2024), retrieved 2026-10-08: https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk
• Gartner, "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027," press release, 25 June 2025, retrieved 2026-10-26: https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027
• Gartner, "Gartner Predicts 70% of Enterprises Will Abandon Agentic AI Built by Vendor Forward-Deployed Engineering by 2028," press release, 29 September 2026, retrieved 2026-10-26: https://www.gartner.com/en/newsroom/press-releases/2026-09-29-gartner-predicts-70-percent-of-enterprises-will-abandon-agentic-ai-built-by-vendor-forward-deployed-engineering-by-2028
• Gartner, "Gartner Says Nearly Half of CIOs Are Planning to Deploy Artificial Intelligence," press release, 13 February 2018, retrieved 2026-10-26: https://www.gartner.com/en/newsroom/press-releases/2018-02-13-gartner-says-nearly-half-of-cios-are-planning-to-deploy-artificial-intelligence
• James Ryseff, Brandon F. De Bruhl, and Sydne J. Newberry, The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed: Avoiding the Anti-Patterns of AI, RAND Corporation, RR-A2680-1, 2024 (interviews with 65 data scientists and engineers, 50 from industry; footnote 13 sources the "more than 80 percent" figure to Jeremy Kahn, Fortune, 26 July 2022; footnote 14 sources the IT comparison to Iavor Bojinov, Harvard Business Review, November–December 2023), retrieved 2026-10-08: https://www.rand.org/pubs/research_reports/RRA2680-1.html
• VentureBeat, "Why do 87% of data science projects never make it into production?," 19 July 2019 (report on a Transform 2019 panel with Deborah Leff of IBM and Chris Chapo of Gap), retrieved 2026-10-26: https://venturebeat.com/ai/why-do-87-of-data-science-projects-never-make-it-into-production/
• BARC, "Global Study Maps the Reality of AI Delivery in 2025," 3 December 2025 (421 organizations), retrieved 2026-10-26: https://barc.com/news/ai-leading-edge-study/
• Salesforce, "Salesforce Announces 2026 Connectivity Report," 5 February 2026, for MuleSoft's 2026 Connectivity Benchmark Report (1,050 IT leaders in nine countries, with Vanson Bourne, October to November 2025), retrieved 2026-10-08: https://www.salesforce.com/news/stories/connectivity-report-announcement-2026/
• Writer, "Key findings from our 2025 enterprise AI adoption report," 18 March 2025 (1,600 US knowledge workers using AI at work, including 800 C-suite executives and 800 employees, surveyed with Workplace Intelligence), retrieved 2026-10-08: https://writer.com/blog/enterprise-ai-adoption-survey/
• Frank F. Xu et al., "TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Tasks," Carnegie Mellon University, arXiv:2412.14161, first submitted 18 December 2024, revised 10 September 2025, retrieved 2026-10-26: https://arxiv.org/abs/2412.14161
• Kung-Hsiang Huang et al., "CRMArena-Pro: Holistic Assessment of LLM Agents Across Diverse Business Scenarios and Interactions," Salesforce AI Research, arXiv:2505.18878, 24 May 2025, retrieved 2026-10-26: https://arxiv.org/abs/2505.18878
• Stanford Institute for Human-Centered Artificial Intelligence, The 2026 AI Index Report, Responsible AI chapter, April 2026, retrieved 2026-10-26: https://hai.stanford.edu/ai-index/2026-ai-index-report/responsible-ai
• Federal Reserve Banks, 2026 Report on Employer Firms: Findings from the 2025 Small Business Credit Survey, March 2026 (fielded 3 September to 14 November 2025; 6,525 firms with 1 to 499 employees; convenience sample, weighted), retrieved 2026-10-08: https://www.fedsmallbusiness.org/reports/survey/2026/2026-report-on-employer-firms
• US Census Bureau, "Large Firms With at Least 20 Employees Biggest AI Users," America Counts story, May 2026 (Business Trends and Outlook Survey, December 2025 to May 2026), retrieved 2026-10-08: https://www.census.gov/library/stories/2026/05/ai-use-businesses.html
• Michael Bloch, Sven Blumberg, and Jürgen Laartz, "Delivering large-scale IT projects on time, on budget, and on value," McKinsey & Company, 1 October 2012 (more than 5,400 IT projects, with the BT Centre for Major Programme Management at the University of Oxford), retrieved 2026-10-08: https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/delivering-large-scale-it-projects-on-time-on-budget-and-on-value
• Mark Keil, Joan Mann, and Arun Rai, "Why Software Projects Escalate: An Empirical Analysis and Test of Four Theoretical Models," MIS Quarterly 24(4), December 2000, pp. 631 to 664 (mail survey of ISACA members; 579 surveys returned), retrieved 2026-10-08: https://scholarworks.gsu.edu/cis_facpub/43

On comparing the figures. The statistics above measure different units (organizations, projects, pilots, use cases, tasks), use different definitions of failure, and come from different years and samples. Some are surveys and some are forecasts or benchmarks. Placing them side by side shows how they differ; they shouldn't be averaged or ranked.

On the Gartner figures from October 2026. The 97% figure is stated in Gartner's release without a sample or method. The 38% over-budget and 85% spend-more figures are quoted from an analyst's keynote remarks in the same release, also without a stated sample.

On McKinsey's 37%. It counts respondents attributing any EBIT impact to AI. The complement isn't a measured "no impact" share.

On the Gartner I&O middle band. Gartner publishes the 28% (fully succeed) and 20% (fail outright) figures. The 52% shown in the chart is the remainder and assumes the categories cover all use cases.

On the MIT funnel. MIT reports 60% investigated, 20% piloted, and 5% successfully implemented as shares of organizations. The "one in four" figure is 5 divided by 20 and assumes the implementing organizations are a subset of those that piloted.

On the IDC figure. IDC's underlying research is not public. The 88% and "4 of every 33" figures are as reported by CIO, which attributes them to IDC research conducted with Lenovo.

On IBM's figures. The 25% and 16% are the surveyed CEOs' own estimates of their AI initiatives, not independently measured project outcomes.

On the Census figures. The Census Bureau reworded its core AI question in November 2025, so earlier rates aren't directly comparable.

On small-business abandonment. No government or central-bank survey we found (the Census Bureau's Business Trends and Outlook Survey, the Federal Reserve's Small Business Credit Survey, and Statistics Canada's Canadian Survey on Business Conditions) measures firms that tried AI and stopped.

On the HVAC example. The six hours a week are illustrative, not a measurement.

---

About Avolis Research Group

Avolis Research Group is Avolis's in-house research practice, focused on how operations-heavy small and mid-sized businesses actually adopt AI. It synthesizes primary economic research, government survey data, and results from real implementations into practical, vendor-neutral guidance.

More about Avolis and how we work · Get in touch