Why 95 Percent of GenAI Pilots Never Pay Off
MIT NANDA's "State of AI in Business 2025" report is one of the most cited studies on enterprise AI adoption. Its central finding explains a lot of what shows up in assessments here.
15 minutes. No pitch. No cost.
A Large, Recent Look at Real Enterprise AI Deployments
The report, published by MIT NANDA in 2025 under the title "The GenAI Divide: State of AI in Business 2025," is based on more than 300 enterprise AI deployments, 52 case studies, and 153 leadership interviews. Its headline finding: about 95 percent of enterprise generative AI pilots fail to deliver a measurable return.
The report's own conclusion is not that the technology is weak. It is that most organizations mismanage adoption -- tools get deployed without the workflow redesign, data readiness, or employee buy-in needed to make them stick.
We did not run this study and are not affiliated with MIT or NANDA. We are citing it because the pattern it documents matches what we consistently find in assessments.
Coverage of the report (Forbes) ->Study Snapshot
Source: MIT NANDA, "The GenAI Divide: State of AI in Business 2025."
What the Report Actually Found
Four findings from the report, in plain language, with what each one tends to mean for a smaller business.
Nineteen out of twenty enterprise GenAI pilots studied failed to produce a measurable financial return. The report attributes this to organizations avoiding the friction -- workflow redesign, retraining, governance -- that separates a demo from a working system.
The report groups the causes of failure into human friction (retraining and verification), organizational friction (workflow and governance resistance), and technical friction (tools that do not retain context or improve over time). All three are addressed by an assessment before implementation.
Partnering with a specialized vendor succeeded roughly twice as often as building a custom solution internally. This is part of why implementation here favors proven platforms (n8n, Make.com, direct API integration) over ground-up custom builds.
About 90 percent of employees studied use personal generative AI tools at work, while only around 40 percent of the same organizations have any officially sanctioned AI subscription. Employees are already finding value; the organization has not caught up with governance or training.
The Gap Is Almost Never the Model
None of these findings are really about AI capability. They are about the workflow, the data, and the people around the tool -- which is exactly what an assessment is built to catch early.
Assess before you buy
If 95 percent of pilots fail without one, the assessment step is not overhead -- it is the difference between joining that statistic and not.
AI Assessment ->Favor proven platforms over custom builds
The report's build-versus-buy gap is reflected directly in how implementation is scoped here -- proven tools, wired to your workflow, over ground-up custom development.
AI Implementation ->Close the shadow AI gap deliberately
If your team is already using AI tools informally, that is a training and governance question, not a "should we adopt AI" question.
Customized AI Training ->