The Pulse
Global enterprise AI spending will hit $2.59 trillion in 2026. Yet four out of five AI projects launched this year will fail to deliver their intended business value. That is not a prediction. It is the documented outcome of 2,400 enterprise AI initiatives tracked by the RAND Corporation through 2025. The enterprise AI ROI 2026 crisis is the defining tension of the current technology cycle: boards keep approving budgets, vendors keep cashing cheques, and the same failure patterns keep repeating with near-mathematical precision.
Core Significance
Why it matters:
- The Spending Paradox Is Real: Gartner confirmed this month that worldwide AI spending will reach $2.59 trillion in 2026, a 47% increase year-on-year. In the same report, Gartner placed AI firmly in the Trough of Disillusionment on its hype cycle, explicitly stating that the improved predictability of ROI must occur before AI can truly be scaled up by the enterprise. Companies are spending more and trusting the results less. That is not a cycle correcting itself. That is a structural failure.
- The C-Suite Is Losing Patience: S&P Global found that 42% of companies scrapped most of their AI initiatives in 2025, up sharply from 17% the year before. IBM’s research puts the number of AI initiatives delivering expected ROI at just 25%. Morgan Stanley found only 21% of S&P 500 companies could cite a measurable AI benefit at all. Three separate research firms, three separate methodologies, and the same conclusion: most enterprise AI spending is not working.
- The Measurement Gap Is the Root Cause: A 2025 MIT Sloan study found that 61% of enterprise AI projects were approved on projected ROI that was never measured after launch. Companies are not failing at AI because the technology does not work. They are failing because they never defined what working would look like.
Deep Context: How We Got Here
The current crisis has a specific starting point. Between 2022 and 2024, the launch of ChatGPT and subsequent enterprise AI tools created a buying panic that bypassed normal capital allocation discipline. IT departments that spent years building business cases for basic software purchases approved generative AI pilots in weeks. The pressure came from the top. CEOs who read about ChatGPT on a Sunday wanted a strategy document by Tuesday.
That pressure produced a wave of AI pilots that shared one characteristic: they were designed to demonstrate capability, not deliver value. A chatbot that impressed in a demo. A document summariser that saved thirty minutes a week. A code assistant that developers used occasionally. Each one technically worked. None of them moved a meaningful business metric.
By 2025, the bill arrived. According to the McKinsey State of AI 2025 report, only 39% of organisations reported EBIT impact at the enterprise level from AI, despite 88% reporting AI use in at least one business function. The gap between deployment and value is not closing. It is widening as organisations add more tools without fixing the underlying measurement and governance failures that caused the first wave to underperform.
Gartner’s January 2026 forecast, confirmed again in its May 2026 AI spending report, captured the mood with unusual directness: AI in 2026 will most often be sold to enterprises by their incumbent software provider rather than bought as part of a new moonshot project. The era of the AI experiment is over. The era of justifying the spend has arrived.
Data Insights
By the numbers:
- $2.59 Trillion: Global AI spending forecast for 2026, per Gartner, a 47% increase year-on-year.
- 80.3%: The percentage of enterprise AI projects that fail to deliver intended business value, per RAND Corporation’s meta-analysis of 2,400 initiatives.
- $547 Billion: The estimated value of 2025’s $684 billion global enterprise AI investment that produced no measurable results by year-end.
- 42%: Share of companies that scrapped most of their AI initiatives in 2025, per S&P Global, up from 17% in 2024.
- 61%: Enterprise AI projects approved on projected ROI that was never measured after launch, per MIT Sloan Management Review 2025.
- 39%: Share of organisations reporting enterprise-level EBIT impact from AI, per McKinsey’s State of AI 2025 survey of over 1,000 executives.
- $7.2 Million: Average cost per failed AI initiative at large enterprises in 2025, per RAND Corporation analysis.
| Failure Category | Share of Failures | Avg Cost | Avg Value | Outcome |
| Abandoned pre-production | 33.8% | $4.2M | $0 | Total loss |
| Completed, underperformed | 28.4% | $6.8M | $1.9M | Negative ROI |
| Ran, never recouped cost | 18.1% | $8.4M | $3.1M | Negative ROI |
| Total failure rate | 80.3% | $7.2M avg | Negligible | Negative ROI |
| Success rate | 19.7% | Varies | 1.7x return | Positive ROI |
The table above maps the full anatomy of the enterprise AI ROI 2026 failure landscape. Note that the 19.7% success group returns 1.7x on average proof the technology works when deployed with discipline.
The Business Case: Three Reasons the Same Projects Keep Failing
The RAND data is not a random distribution of bad luck. The failures cluster around three repeatable patterns, each with a specific organisational cause.
Reason 1: No Definition of Success Before Launch
RAND found that 73% of failed AI projects had no agreed definition of success before the project started. This is not a technology failure. It is a governance failure dressed as an AI project. When a chief digital officer approves a pilot to explore AI’s potential in customer service, they have created a project that can never officially fail because it was never officially supposed to succeed at anything specific.
The 19.7% of projects that delivered ROI shared one characteristic before anything else: a specific, measurable outcome agreed upon before the first line of code was written. Not improve customer satisfaction. Not reduce operational friction. A number. A timeline. A named executive accountable for it.
Reason 2: Building on Inadequate Data Foundations
Gartner’s 2025 research found that 60% of AI projects lacking AI-ready data will be abandoned by the end of 2026. Most enterprise data environments were not built for AI. They were built for reporting. The difference matters enormously. Reporting systems store historical data in structured formats designed for human analysts. AI systems need clean, labelled, connected data designed for model training and inference. Most large enterprises have years of data that is technically present and practically useless for AI.
Companies that skipped the data foundation phase are now paying twice: once for the AI tools that underperformed, and again for the data remediation work that should have happened first.
Reason 3: Treating AI as IT Projects Rather Than Business Transformation
RAND identified this as the single highest-correlation factor with failure. When an AI project is owned by IT, its success metric is deployment. When it is owned by the business function it is supposed to transform, its success metric is value delivered. The distinction sounds administrative. The outcomes are dramatically different.
Projects with sustained CEO involvement achieve a 68% success rate versus 11% for those that lose C-suite sponsorship within six months, per RAND’s analysis. The technology is not the variable. Leadership attention is the variable.
Between the lines:
The companies winning on enterprise AI ROI 2026 are not the ones with the most sophisticated models. They are the ones that treated AI adoption as a change management programme with a technology component, not a technology programme with a change management afterthought. Microsoft, Salesforce, and ServiceNow are all reporting strong enterprise AI attachment rates not because their models are superior to OpenAI’s but because they embedded AI into workflows people already used every day. The bar for adoption dropped to zero because the switch was already flipped.
Regional Spotlight: Pakistan’s SMEs and the AI ROI Trap
The enterprise AI failure crisis playing out in Fortune 500 boardrooms is arriving in Pakistani businesses with one critical difference: the budgets are smaller and the margin for error is zero.
The Opportunity:
Pakistan’s AI market reached $949 million in 2025 and is projected to grow to $3.23 billion by 2030, according to Statista data cited by Invest2Innovate. A recent Kaspersky survey found that 86% of professionals in Pakistan report using AI tools for work tasks, a figure higher than many developed nations. Pakistani SMEs in fintech, healthtech, and e-commerce are adopting AI tools at a pace their infrastructure would not suggest is possible.
The opportunity is real. The problem is that Pakistani enterprises are importing the same failure patterns that destroyed ROI for companies with ten times their budget. A Lahore-based logistics firm that deploys an AI route optimiser without measuring its baseline delivery costs first has made the same mistake as a US retailer that spent $40 million on a generative AI platform without defining what success looked like.
The Crisis:
Pakistan’s National AI Policy targets nationwide AI awareness for internet users by 2026 and one million trained learners by 2027. These are capability targets. There are no published targets for enterprise AI ROI, for successful deployment rates, or for the percentage of AI investments generating measurable economic return. Pakistan is building a generation of AI users without simultaneously building a framework for AI accountability. As the Siri 2.0 article on The Brief Script highlighted, even the most sophisticated AI integrations fail when the governance layer is missing.
The skills gap will close. The measurement gap, without deliberate intervention, will not. Pakistan’s $1 billion National AI Fund is a financing instrument. It is not, by design, a governance framework. That gap needs its own policy response before the fund’s first major deployments reach their review cycle.
Expert Nuance: The Real Crisis Is Not Failure. It Is Invisibility.
Every headline about enterprise AI failure focuses on the 80.3% that did not deliver. The more important number is the 61% whose failure was never officially recorded because ROI was never measured in the first place.
An AI project that was approved, deployed, quietly underperformed, and was eventually replaced by a newer tool does not appear in failure statistics. It appears in the next budget cycle as a legacy system replacement. The actual cost of enterprise AI underperformance is significantly higher than the RAND figure suggests because RAND only captures documented failures. The invisible failures, the ones where the tool ran but the expected productivity gains never materialised, are the majority.
This is why the companies with the strongest enterprise AI ROI 2026 results are investing as much in measurement infrastructure as in AI infrastructure. Citi identified a 30 basis point credit spread penalty for companies classified as AI adopters versus enablers, meaning debt markets are already pricing in the difference between spending on AI and proving it works. The gap between adoption and proof is now priced into the cost of capital. As the Nvidia AI chip competition analysis on The Brief Script noted, the companies controlling infrastructure are compounding their advantage the same compounding effect applies to governance.
Boards that have not noticed this yet will notice when their next refinancing round prices the difference.
Strategic Outlook: What’s Next
Three forces will define how the enterprise AI ROI 2026 story develops over the next 18 months.
- The Governance Wave: The next 18 months will see enterprise AI investment shift from model procurement toward AI governance infrastructure. Measurement platforms, accountability frameworks, and AI auditing tools will be the fastest-growing segment of the enterprise AI market through Q4 2026 and into 2027. Companies like Workday, SAP, and Oracle are already embedding governance dashboards into their AI product suites. This is not a compliance trend. It is a ROI recovery trend.
- Agentic AI as the Reset Button: The shift from generative AI tools to agentic AI workflows gives enterprises a structural opportunity to rebuild their AI strategy on a measurable foundation. Unlike a chatbot that produces outputs with no direct connection to business outcomes, an AI agent that executes a multi-step procurement workflow produces a measurable transaction record. Every action is logged. Every outcome is attributable. The companies that struggled to measure GenAI ROI will find agentic AI considerably easier to justify, provided they define success metrics before deployment.
- The Vendor Accountability Shift: Microsoft, Google, and Salesforce are all moving toward outcome-based pricing models for enterprise AI in 2026 and 2027. Instead of selling seats and licences, they will increasingly tie pricing to measurable outcomes: cost per resolved ticket, revenue per AI-assisted sales interaction, time saved per automated workflow. This shift transfers accountability from the enterprise AI team to the vendor for the first time, and it will accelerate adoption among enterprises that failed in the first wave and are cautious about committing new budget.
Key Question Answered
Why are enterprise AI projects failing to deliver ROI in 2026 and what can companies do differently?
According to RAND Corporation’s analysis of over 2,400 enterprise AI initiatives, 80.3% of projects fail to deliver their intended business value in 2026. The three primary causes are: launching without a specific, measurable definition of success; building AI systems on inadequate data foundations; and treating AI as an IT project rather than a business transformation programme. The 19.7% of projects that succeed consistently share three characteristics before launch: a specific numeric outcome with a deadline, executive sponsorship sustained beyond the pilot phase, and a data readiness assessment completed before model deployment. Gartner’s 2026 forecast confirms that improving enterprise AI ROI 2026 requires shifting investment from AI tools toward AI governance and measurement infrastructure.
The Takeaway
The $2.59 trillion being spent on AI in 2026 is not being wasted on bad technology. It is being wasted on good technology deployed without the organisational discipline to prove it is working. Every failed AI project in the RAND data was technically functional. The tools ran. The models generated outputs. The failure happened at the layer of human decision-making: the executive who approved a pilot with no success criteria, the data team that was never asked if the data was ready, the IT manager who declared the deployment a success when the contract was signed rather than when the business metric moved. The technology is not the problem. The governance around it is. Companies that understand this distinction in 2026 will be the ones writing the success stories in 2027.
