AI Transformation · Adaptability
The AI ROI Gap
Why most enterprise AI investment never reaches financial returns.
Enterprises have spent tens of billions of dollars on generative AI and most of it has produced no measurable return. The evidence points to a cause that has nothing to do with the models, and everything to do with the organization around them.
of enterprise GenAI initiatives show no measurable P&L return
MIT NANDA, State of AI in Business 2025
The spend is real. The return depends on organizational conditions, not the model.
Diagnose your constraint →Research Foundation
- Peer-reviewed organizational and cognitive research
- AI and technology research (MIT, Stanford HAI)
- Executive leadership literature
- Enterprise transformation field experience
- The Builders Build Framework
Key Takeaways
- Roughly 95 percent of enterprise generative AI pilots deliver no measurable profit-and-loss impact, despite tens of billions in spending (MIT Project NANDA, 2025).
- The binding constraint is organizational learning, not the technology. Value appears only when the organization changes how work is actually done and how decisions are made.
- The Four A's give executives a diagnostic for the conditions ROI depends on: Attention to concentrate effort, Alignment on where value is, Authority to redesign the work, and Adaptability so the change survives.
What the evidence actually shows
In 2025, MIT’s Project NANDA published The GenAI Divide: State of AI in Business 2025. Drawing on a review of more than 300 publicly disclosed AI initiatives, 52 structured executive interviews, and 153 survey responses, it reported that despite an estimated 30 to 40 billion dollars in enterprise investment, about 95 percent of generative AI pilots produced no measurable return. Only around 5 percent reached meaningful revenue or cost impact.
This is not an isolated statistic. Stanford’s AI Index has tracked a widening gap between AI adoption and AI value realization, and industry research from firms such as BCG and Deloitte has repeatedly found that a small minority of organizations capture the large majority of returns. The pattern is consistent enough to treat as a finding rather than a headline: the technology works, and most organizations still do not get paid for it.
The cause is a learning gap, not a technology gap
MIT’s central explanation is worth quoting precisely: the core barrier is learning, not infrastructure, regulation, or talent. Most deployments do not retain feedback, adapt to context, or change the surrounding process, so the organization never absorbs the capability. This is a very old finding wearing new clothes. Argyris and Schön showed decades ago that organizations improve only when they change the assumptions and routines that govern behavior, not merely their tools. Senge described the same dynamic as the difference between organizations that learn and organizations that merely install.
Technology adoption research says the same thing from a different angle. Davis’s Technology Acceptance Model established that usage depends on perceived usefulness and ease of use inside real work, not on the sophistication of the tool. A model that is never woven into how decisions are made produces activity, not value.
The organizations that capture AI value are not the ones with the best models. They are the ones that were able to change how work is done.
Why more capability can lower returns
There is a cognitive reason the naive answer, buy more AI, backfires. Simon’s work on bounded rationality established that executive attention, not information, is the scarce resource in decision-making. Sweller’s research on cognitive load and Kahneman’s synthesis of dual-process reasoning show that adding inputs beyond a person’s working capacity degrades judgment rather than improving it. Layering more AI outputs onto an organization that has not decided where value is, or who may act on it, increases load without increasing throughput.
MIT’s own breakdown is telling. More than half of generative AI budgets went to sales and marketing tools, while the largest measured returns came from back-office automation, and externally partnered deployments succeeded roughly twice as often as internal builds. In other words, value tracked with focus and with a willingness to change the operating model, not with spending.
The Four A’s reading of the ROI gap
The research establishes the phenomenon. The Four A’s of Organizational Readiness provide a practical executive model for recognizing the conditions ROI depends on, and for seeing which one is missing before the next investment is approved.
Attention determines whether effort is concentrated where value actually is, rather than scattered across dozens of pilots. Alignment determines whether the organization shares a real definition of the value it is chasing, or only agreed to it in a meeting. Authority determines whether anyone is empowered to redesign the work and the decision rights the tool implies. Adaptability determines whether the change survives contact with the organization, or reverts the moment attention moves on. The MIT learning gap is, in this language, an Adaptability failure sitting on top of an Attention and Authority failure. That is why the ROI does not arrive: the organization installed a capability it was not structured to absorb.
Evidence matrix
Each claim below separates what the research establishes, what field experience observes, and what the Four A’s interpret. The framework is a synthesis built on established science, not a replacement for it.
| Claim | Research | Field evidence | Four A’s |
|---|---|---|---|
| Most AI pilots produce no measurable return | MIT Project NANDA (2025); Stanford HAI AI Index | Stalled enterprise modernizations | Adaptability |
| Value requires changing how work is done | Argyris & Schön (1978); Senge (1990); Davis (1989) | A federal data modernization | Authority |
| More inputs beyond capacity lower decision quality | Simon (1947); Sweller (1988); Kahneman (2011) | Executive review overload | Attention |
| Returns concentrate where focus and ownership are clear | MIT NANDA back-office and buy-vs-build findings | Honeywell operating conditions | Alignment |
What executives should do before the next AI investment
Do not approve another pilot on the strength of the demo. Ask which organizational condition would have to be true for this capability to reach the P&L, and whether it is. Concentrate spending on a small number of high-value processes rather than many. Name the person with the authority to redesign the work the tool implies. And require a plan for how the change will be learned and retained, not just launched. The model is rarely the constraint. The organization around it almost always is.