What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
The often-cited 95% figure is not a verified failure rate for every company AI project. It comes from the 2025 GenAI Divide report associated with MIT’s Project NANDA; Maynooth University describes it as initiatives that had no positive or negative impact on their organization after deployment. That is a narrower claim than “95% of AI projects fail,” and the accessible account does not establish the original report’s full methodology or denominator. The more useful question is why pilots often fail to become measurable, organization-wide value.
What does the 95% figure actually measure?
Maynooth University attributes the headline to the 2025 GenAI Divide: State of AI in Business report and characterizes the reported outcome as AI initiatives having “no positive or negative impact on their organisation after deployment.” That describes a lack of reported impact, not necessarily a project that was technically broken, abandoned, or harmful.
The original report PDF linked by Maynooth was not accessible for independent review, so its precise definition, sampling, and denominator cannot be confirmed from the available account. The figure should therefore be attributed to that report rather than presented as a representative census or a settled rate for all corporate AI. It is also specifically about generative AI initiatives, not every kind of AI system a company might deploy. Maynooth University’s summary is the accessible account of the claim.
Why do AI pilots stall before they create value?
A successful demonstration is not a scale decision
A pilot can perform well on a narrow task and still have little organizational impact. It may remain isolated, have no owner or budget for its next phase, or produce lessons that are never reused. Without agreed criteria for what counts as success—and a decision point to stop, revise, or scale—experimentation can continue without changing how work gets done.
#1 Best Overall
Maynooth illustrates this with an Irish public-sector organization that built a legislation-review prototype and a separate chatbot for citizen queries. Both had limited positive outcomes, but leadership deferred broader scaling. The example shows how separate experiments can accumulate as disconnected tools rather than becoming an integrated capability; it is illustrative, not a measure of how often this happens across industry.
Weak data and processes limit what a model can do
Fragmented data, outdated systems, and workflow bottlenecks make it difficult to deliver reliable outputs where employees need them. Placing an AI layer over a confusing or broken process may automate some steps without fixing the underlying work. Process simplification, data preparation, and integration need to be part of deployment planning—not deferred until after a pilot appears successful.
Governance and workforce readiness are not optional extras
In a 2025 TDWI survey, respondents frequently named lack of governance, lack of AI literacy, and hallucinations as frustrations. These are reported concerns, not proof that any one of them caused the NANDA result. They do help explain why an output that looks promising in a demo may be difficult to trust or use in routine work.
Employees need to know when AI is appropriate, how to check its outputs, and where to raise concerns. Organizations also need rules for data use, oversight, and accountability. As Tamilla Triantoro, professor of business analytics at Quinnipiac University and TDWI contributor, put it: “If you aren’t prepared to redesign your workflows, govern your data, and upskill your people, you aren’t driving transformation—you’re accelerating inefficiency.”
Rank #2
What other evidence says about company adoption
TDWI’s 2025 survey offers a separate view of business use, not a replication of the 95% figure. It was fielded in June and July 2025 across company sizes, industries, and roles; more than 200 people participated, and 155 responses met the stated quality criteria. Its results describe those respondents and should not be treated as universal benchmarks.
| TDWI survey finding | What it means |
|---|---|
| 90% of respondents | Reported already using general-purpose GenAI assistants. |
| 49% of respondents | Cited lack of governance as a frustration. |
| 48% of respondents | Cited lack of AI literacy as a frustration. |
| 46% of respondents | Cited hallucinations as a frustration. |
| Among “builder” organizations, 64% | Reported faster decision-making. TDWI defines builders as organizations using GenAI with their own data to create applications and operational workflows. |
| Among “builder” organizations, 46% | Reported increased innovation, using the same builder definition. |
The frustration percentages are not mutually exclusive: a respondent could report more than one issue. TDWI vice president of research Fern Halper summarized the scale challenge this way: “MIT is right that scaling AI pilots for measurable ROI is challenging.” TDWI’s survey summary provides its field dates and respondent details.
Scaling is an organizational capability, not a model upgrade
MIT Sloan’s account of MIT CISR research presents four stages: experiment and prepare; build pilots and capabilities; industrialize AI throughout the enterprise; and become AI future-ready. The framework describes a progression organizations can use to assess their work, not a guarantee that every company advances in the same order.
Recommended Free Tools
The article draws on a 2022 MIT CISR survey of 721 companies and nine enterprise executive interviews conducted in 2024. In the 2022 survey, 28% of companies were in stage one, 34% in stage two, 31% in stage three, and 7% in stage four. Those proportions describe that survey, not a current global distribution.
Rank #3
Experiment and prepare
Identify a genuine business or user problem, prepare data, and establish outcomes that can be measured. This stage is about choosing useful opportunities and understanding what must change before deployment.
Build pilots and capabilities
Test use cases while developing the skills, governance, metrics, and technical foundations required to learn from them. A pilot should generate evidence about the work and the operating model, not only show that a model can produce an output.
Industrialize across the enterprise
Scaling calls for dependable architecture, prepared data, transparent outcomes, and processes that can support broader use. It also requires coordination across teams so each project does not reinvent controls and infrastructure.
Free tools Windows power users keep installed
One-click scans. No signup required.
Become AI future-ready
At the most advanced stage in the framework, AI is supported by organization-wide capabilities and continuing adaptation. MIT Sloan’s account emphasizes that reaching this point involves sustained organizational change, rather than a one-time technology rollout.
Peter Weill, MIT CISR chair, told MIT Sloan: “The hardest part of Stage 2 is changing.” He also emphasized simplifying and automating processes before applying AI to them. MIT Sloan’s overview of the MIT CISR maturity model details the framework and its research basis.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to decide whether an AI project should scale
Before moving from a pilot to broader deployment, assess the project against the work it is meant to improve. A useful decision should make it possible to stop or revise a weak use case as confidently as it can justify investment in a strong one.
- Value evidence: Define the user or business outcome in advance and measure it in real work, not only in a demonstration.
- Workflow fit: Identify which steps will change, who will use the system, and whether the current process needs simplifying first.
- Data readiness: Check that required data is accessible, suitable, and connected to the systems where work happens.
- Governance and risk: Set expectations for oversight, data handling, output checking, and accountability.
- People and adoption: Provide training and support, and include employees who understand the workflow in redesign decisions.
- Scale path: Name the owner, integration needs, reusable capabilities, and resources required to expand—or the evidence that would trigger a stop.
Set pilot exit criteria before work begins. At the decision point, choose to scale when evidence supports the intended outcome and the organization can operate the solution responsibly; revise when a fixable gap remains; or stop when value is not demonstrated. Capture what the pilot teaches so future teams can reuse useful components and avoid repeating disconnected experiments.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
