AI is making individuals measurably faster and better at their work. So why is almost none of that showing up on the bottom line? The latest research reveals a gap that should concern every leader responsible for business transformation.
AI has crossed a threshold in the workplace. The question for leaders is no longer whether people are using it, but why the gains showing up in individual performance are so stubbornly failing to show up on the bottom line.
The evidence is now clear enough to be uncomfortable. Employees feel more productive. Executives report transformation. Yet the enterprise-level financial impact of AI has, by the most rigorous measure available, barely moved in two years.
This is the productivity paradox of the current AI wave, and understanding it is the single most important task facing any leader responsible for business transformation today.
The Individual Wins Are Real
Start with what is unambiguously working. In McKinsey's 2026 State of AI global survey of 1,993 participants, 80% of respondents report that AI has improved their individual productivity, and 50% say it helps them make better decisions. (McKinsey, 2026)
These are not marginal effects. They are consistent across organisational levels, from individual contributors up to senior executives. Deloitte's 2026 State of AI in the Enterprise report found that two-thirds (66%) of organisations report productivity and efficiency gains from AI, and that worker access to AI tools rose by 50% in 2025 alone. (Deloitte, 2026)
The most controlled evidence comes from a landmark field experiment run by Harvard Business School, MIT, and the Wharton School with 758 BCG consultants. Consultants using GPT-4 completed 12.2% more tasks, 25.1% faster, and produced results rated over 40% higher in quality than those working without AI. (Brynjolfsson, Li & Raymond, 2023)
At the individual level, the case is closed. AI makes knowledge workers faster and, on the right tasks, better.
The Enterprise Gap
Here is the puzzle. If individuals are demonstrably more productive, why is that not translating into enterprise performance?
The McKinsey data is striking. The share of organisations reporting that AI has contributed to their EBIT (earnings before interest and taxes) has been essentially flat at 37%, unchanged from 2025. The proportion of genuine AI high performers, those attributing at least 5% of EBIT to AI, has remained stuck at roughly 6% of all respondents. (McKinsey, 2026)
Deloitte's research points to the same gap from a different angle. While twice as many leaders as last year report transformative impact, only 34% of organisations are truly reimagining their business around AI. The majority are using AI at the surface level or redesigning isolated processes, not transforming how the organisation actually works. (Deloitte, 2026)
The pattern is consistent across every major study: individual productivity is up, but enterprise value is not following. This is not a technology problem. It is a transformation problem.
The Skill Atrophy Problem
One reason the gains fail to compound is that AI may be quietly eroding the very capabilities that produce them.
A 2025 study from Microsoft Research and Carnegie Mellon University, surveying 319 knowledge workers, found that higher confidence in AI is associated with less critical thinking. When workers trust the tool, they offload the cognitive work of evaluating, questioning, and improving its output. The study's authors describe a shift from creating to reviewing, and warn of long-term over-reliance and diminished skill. (Lee et al., CHI 2025)
This matters because reviewing work you did not produce, at the volume AI enables, is cognitively exhausting in ways that doing the work yourself is not. It is the same dynamic Lisanne Bainbridge identified in her 1983 paper Ironies of Automation: the more reliable the automation, the less engaged, and therefore the less capable, the human operator becomes.
The implication for transformation is serious. If AI adoption erodes the judgment and critical thinking of the people who are supposed to be driving the transformation, the organisation becomes more dependent on a tool it understands less and less. That is not a foundation for lasting change.
The Overconfidence Trap
The Harvard and BCG experiment revealed a second, equally important finding. On tasks that fell outside AI's "jagged frontier" of competence, the consultants using AI were more likely to produce confident but incorrect solutions. The tool did not just fail to help on these tasks; it actively misled users into trusting outputs that were wrong. (Brynjolfsson, Li & Raymond, 2023)
The Microsoft study found the same psychological mechanism at work. Confidence in AI is the variable most strongly associated with reduced critical thinking. The people most at ease with the tool are the ones most likely to accept its output without scrutiny.
This is the overconfidence trap, and it has a direct bearing on transformation. Leaders who feel productive using AI begin to trust their own AI-assisted analysis more. Decisions get made faster. But faster is not the same as better, and the people making the decisions are often the least well placed to judge where the AI is confident and wrong.
Why Transformation Stalls
Put these findings together and the enterprise gap starts to make sense.
Individual productivity gains do not become enterprise value on their own. They only translate when the organisation changes around them: when workflows are redesigned, when roles are restructured, when measurement and incentives shift to reward the new way of working.
The McKinsey report is explicit on this. The organisations that translate individual gains into enterprise-level performance are those that transform their businesses, not just adopt AI tools. Half of the AI high performers intend to use AI to fundamentally transform their organisations, and most are redesigning workflows as a result. (McKinsey, 2026)
Most organisations are not doing this. They are layering AI onto existing structures and expecting the technology to do the transforming on its own. It will not. Technology adoption without organisational redesign produces efficiency gains that are real but trapped, absorbed by the friction of processes, hierarchies, and incentives that were built for a previous era.
There is also a workforce dimension. McKinsey found that 39% of respondents now expect AI-related declines in employment in the coming year, up from 32% last year. Yet only 14% reported actual declines over the past year, less than half the share who expected them. (McKinsey, 2026) The gap between expected and actual workforce change reveals how much of the current narrative is driven by assumption rather than evidence, and how destabilising that uncertainty is for the people expected to deliver the transformation.
What This Means for Leaders
The research points to a clear set of implications for anyone leading a business transformation in which AI plays a part.
Redesign the work, not just the tools. The single biggest predictor of enterprise value is whether the organisation is willing to change how it operates. Layering AI onto unchanged processes produces individual gains that never reach the bottom line.
Protect critical thinking. The evidence that AI erodes the very cognitive capabilities it depends on is now strong enough to act on. Build verification, peer review, and deliberate periods of unassisted work into AI-assisted workflows. Treat critical thinking as a capability to be developed, not an overhead to be eliminated.
Design for the jagged frontier. AI is brilliant on some tasks and misleading on others. Leaders need to know which is which inside their own organisation, and build decision processes that do not collapse when the tool is confidently wrong.
Close the perception gap. The gap between executives who believe AI is delivering and workers who experience something different is itself a risk. Measure the actual experience of AI use across levels, not just the headline adoption figures.
Treat workforce change as evidence, not assumption. Expectations of AI-driven job losses are running well ahead of reality. Decisions made on assumption rather than evidence destabilise the workforce and undermine the trust transformation requires.
The deepest lesson in the data is this. AI is genuinely transforming what individuals can do. But the leap from individual capability to organisational transformation has never been, and will never be, made by a tool. It is made by leaders willing to do the harder, less glamorous work of changing the organisation around the technology.
The organisations that will capture the real value of AI are those that bring the same discipline to redesigning how they work as they bring to deploying the technology itself. The research is unambiguous about which approach produces results, and which produces another year of flat enterprise impact.
Sources
- McKinsey, The State of AI in 2026: On the Road to ROI (global survey of 1,993 participants, June, July 2025). mckinsey.com
- Deloitte, State of AI in the Enterprise, 2026 AI Report. deloitte.com
- Lee, H., Sambasivan, N., et al., The Impact of Generative AI on Critical Thinking, Microsoft Research and Carnegie Mellon University, CHI 2025. microsoft.com
- Brynjolfsson, E., Li, D., & Raymond, L., Navigating the Jagged Technological Frontier, Harvard Business School, MIT, and Wharton, 2023 (field experiment with 758 BCG consultants). hbs.edu
Matthew Bayfield