The 2025 reality check
Three findings from 2025 landed hard. A preliminary MIT NANDA study reported that roughly 95% of organisations were seeing no measurable profit-and-loss return from generative AI[1]. S&P Global found the share of companies abandoning most of their AI initiatives had jumped to 42%, up from 17% a year earlier[2]. Gartner predicted more than 40% of agentic AI projects would be cancelled by the end of 2027[3].
Set against adoption, the contrast is the whole story. McKinsey found 88% of organisations were using AI regularly by late 2025[5]. Almost everyone is doing it. Almost no one can point to the earnings line and show you where it landed.
Why the gap is real even if 95% is arguable
The 95% figure deserves a caveat, and we will give it one: that study was preliminary, not peer-reviewed, used a short measurement window, and drew criticism on its sample[1]. Take it as directional, not gospel. The point survives the caveat, because the other evidence is independent and consistent. S&P’s abandonment data, Gartner’s cancellation forecast, and BCG’s finding that 74% of companies struggle to scale AI value all say the same thing[2][3][4].
And they all point the same direction for the cause. RAND’s analysis of failed AI projects put the blame on misunderstood requirements and poor data, not on the model[6]. BCG frames success as roughly 70% people and process, 20% technology, 10% algorithms[4]. The technology is rarely what breaks.
The learning gap
The NANDA report’s more useful contribution was not the headline number but the diagnosis. It found that pilots stall when the tool does not learn from feedback or does not fit the way people already work. A demo that dazzles in a controlled setting becomes shelfware the moment it meets a real workflow with real exceptions. Notably, the report found bought-in solutions and genuine partnerships tended to outperform ambitious internal builds, because they crossed that workflow gap more often.
What the minority who succeed do differently
The organisations that get a return are not using better models. They are more disciplined about a short list of things.
- They pick one high-value use case and define the business metric before they evaluate any tool
- They fix the data and the workflow the AI has to live inside, rather than bolting AI onto a broken process
- They put a named owner on the outcome, not just the technology
- They measure against the metric they set, and they are willing to stop a pilot that misses it
This is the same discipline that separates the successful fifth from the failing majority in the broader AI record, covered in our companion guide on why 80% of AI projects fail. The numbers change; the cause does not.
The board’s job
For a board, the trap is being reassured by the wrong metric. Adoption is easy to report and feels like progress: seats deployed, prompts run, staff trained. None of it is return. Ask instead for the profit-and-loss impact of each material use case, and fund the unglamorous work, the data and the process redesign, that decides whether the tool ever pays back.