What is the real AI project failure rate?

There are several different figures because they measure different things. RAND (2024) found more than 80% of AI projects fail — roughly twice the rate of non-AI IT projects. MIT NANDA (2025) found 95% of enterprise generative-AI pilots show no measurable P&L return. Gartner predicts around 30% of generative-AI projects are abandoned after proof of concept by end of 2025. S&P Global (2025) found 42% of businesses scrapped most of their AI initiatives in 2025, up from 17% in 2024. Each measures a different population over a different scope.

Why do AI projects fail so often?

The common thread is not the model — it's people and process. BCG's 10/20/70 framing holds that roughly 70% of AI success depends on people, process and change management, only about 10% on the algorithm. Projects fail because organisations buy a model, plug it in, and never redesign workflows, retrain staff, fix the underlying data, or define how success is measured. Poor data readiness, unclear use cases, and no adoption plan are the recurring causes.

What is the difference between the MIT 95% and RAND 80% figures?

They are not the same claim. RAND's >80% (2024) is about AI projects broadly — and notes that's roughly twice the failure rate of non-AI IT projects. MIT NANDA's 95% (2025) is narrower and stricter: it's about enterprise generative-AI pilots showing no measurable profit-and-loss impact. Different scope, different definition of 'failure,' different year. Quoting them as if they're one number is a credibility mistake.

How does an AI consultant prevent project failure?

By front-loading the parts that actually cause failure. That means: validating data readiness before building, choosing a use case with a clear P&L link, designing the workflow and adoption plan (the 70% in BCG's 10/20/70), defining a success metric up front, and setting a kill-switch. The model is rarely the hard part. Most prevented failures are prevented in scoping, not in engineering.

Why Do AI Projects Fail? The Real Numbers (2025–2026)

Q: Do 95% of AI projects fail?

No — that is a misquote. The accurate statement is that 95% of enterprise generative-AI pilots show no measurable P&L impact, according to MIT's NANDA initiative report 'The GenAI Divide: State of AI in Business 2025' (August 2025). That is specifically about custom enterprise pilots, not all AI and not individual ChatGPT use, which generally works. Saying '95% of AI fails' is wrong and overstates the case.

The widely-cited AI failure statistics — MIT 95%, RAND 80%, Gartner 30%, S&P 42% — explained precisely. They measure different things. Plus why projects fail and how to prevent it.

No — 95% of AI projects do not fail. That number is one of the most misquoted statistics in business. The accurate version is narrower and more useful: 95% of enterprise generative-AI pilots show no measurable P&L impact, per MIT's NANDA report The GenAI Divide: State of AI in Business 2025 (August 2025). The headline failure statistics floating around — 95%, 80%, 42%, 30% — are real, but they measure different things. Understanding the difference is the first step to not becoming one of them.

This post does two things: it states each number precisely with its source, and then explains the actual why — and how a consultant prevents it.

The Four Numbers, Stated Correctly

Figure	What it actually measures	Source
95%	Enterprise generative-AI pilots with no measurable P&L impact	MIT NANDA, State of AI in Business 2025 (Aug 2025)
Over 80%	AI projects that fail — roughly twice the rate of non-AI IT projects	RAND, RRA2680-1 (2024)
42%	Businesses that scrapped most AI initiatives in 2025 (up from 17% in 2024)	S&P Global Market Intelligence (2025)
~30%	Generative-AI projects abandoned after POC by end of 2025 (a prediction)	Gartner (2024)

These are four different populations, four different definitions of "failure," and three different years. Treating them as one number — "AI fails 95% of the time" — is wrong and damages your credibility the moment someone checks.

The MIT 95%: No P&L Impact From Enterprise Pilots

MIT's NANDA initiative published The GenAI Divide: State of AI in Business 2025 on 18 August 2025. Its central finding: 95% of enterprise generative-AI pilots show no measurable return to the profit-and-loss statement. Only around 5% succeed. The research drew on roughly 150 interviews, 350 survey responses and 300 deployments.

Two critical caveats the headlines drop:

This is about enterprise/custom pilots — bespoke GenAI projects inside organisations — not individual tool use. Personal ChatGPT use generally works. The "divide" in the title is exactly that gap: individuals get value, enterprise pilots stall.
"No P&L impact" is a strict bar. A pilot can produce useful outputs and still count here if it never moved a financial metric.

So the honest takeaway is not "AI doesn't work." It's "most enterprise GenAI pilots, as currently run, don't reach the P&L."

The RAND >80%: Broad AI Project Failure

RAND Corporation's 2024 report The Root Causes of Failure for Artificial Intelligence Projects (RRA2680-1) found that more than 80% of AI projects fail — roughly twice the failure rate of comparable non-AI IT projects.

This is broader than MIT's figure. It covers AI projects generally, not just generative-AI pilots, and its benchmark is telling: AI is about twice as likely to fail as ordinary IT. That comparison is the useful part — it says the problem is specific to how AI projects are scoped and run, not just general project risk.

(Note: some write-ups cite precise decimal breakdowns of the causes. Treat those specific percentages as unverified — the safe, citable RAND claim is ">80%, roughly twice non-AI IT.")

The S&P 42%: A Sharp Year-on-Year Trend

S&P Global Market Intelligence reported in 2025 that 42% of businesses scrapped most of their AI initiatives — up sharply from 17% in 2024. On average, organisations abandoned 46% of their proof-of-concepts.

The value here is the trend, not just the level. Abandonment more than doubled year on year. That suggests a wave of 2023–2024 experimentation hitting reality in 2025 — companies pulling back after pilots failed to deliver.

The Gartner ~30%: A Prediction, Not a Measurement

Gartner predicted that around 30% of generative-AI projects will be abandoned after proof of concept by end of 2025. This is a forecast, not an observed result — and it's important to keep Gartner's predictions separate from each other:

~30% of generative-AI projects abandoned after POC by end of 2025.
Over 40% of agentic AI projects cancelled by end of 2027.
60% of AI projects unsupported by AI-ready data abandoned through 2026.

These are three distinct predictions about three different things (GenAI POCs, agentic AI, data-readiness). They are routinely conflated. They shouldn't be.

So Why Do AI Projects Actually Fail?

The numbers describe the what. The why is remarkably consistent — and it's not the model.

BCG's 10/20/70 framing captures it: roughly 10% of AI success is the algorithm, 20% is technology and data, and 70% is people, process and change management (attributed to BCG's Ernesto Pagano). Projects fail because organisations spend on the 10% and skip the 70%:

No clear P&L link. The pilot produces outputs nobody monetises — the MIT 95%.
Data not ready. The model is fine; the data feeding it is incomplete, inaccessible or dirty — the Gartner 60%.
No adoption plan. Staff workflows never change, so the tool sits unused.
No success metric. Without a defined target, "did it work?" is unanswerable — so it defaults to "no measurable impact."
Scope mismatch. Ambitious custom builds where a simpler tool would have delivered.

How a Consultant Prevents It

Most AI failures are prevented in scoping, not engineering. The model is rarely the hard part. A consultant earns their fee by front-loading the parts that cause failure:

Pick a P&L-linked use case — one where success shows up in revenue, cost or time saved.
Validate data readiness first — before building anything.
Plan the 70% — workflows, owners, training, incentives.
Define success and a kill-switch — the metric that means "worked," and the condition to stop.

For a UAE service business, this usually means starting smaller and more boring than the hype suggests — one workflow, one measurable outcome — rather than a sweeping "AI transformation" that lands in the 95%.

We cover the framework behind this in 70/20/10 for AI and the full delivery model in the AI Consulting in Dubai pillar guide. To pressure-test a specific AI initiative before you spend on it, book an AI consultation via evolvxai.com or read more about how we work.

Sources

MIT NANDA — The GenAI Divide: State of AI in Business 2025 (Aug 18 2025), 95% of enterprise GenAI pilots no P&L return: finance.yahoo.com; report mirror mlq.ai
RAND Corporation — The Root Causes of Failure for AI Projects (RRA2680-1, 2024), >80% fail, ~2x non-AI IT: rand.org
Gartner — 30% of generative-AI projects abandoned after POC by end of 2025: gartner.com; over 40% of agentic AI projects cancelled by 2027: gartner.com
S&P Global Market Intelligence (2025) — 42% scrapped most AI (up from 17%), 46% POC abandonment: ciodive.com
BCG / Ernesto Pagano — 10/20/70 people-centric AI: pmi.org/blog/ai-transformation-people-insights-bcg