AI is everywhere. So where is the value?
A version of the same conversation has been happening in more than one client room lately.
The conversation starts with AI and, very quickly, the list gets long. Copilot has been rolled out. There are dozens, sometimes hundreds, of AI use cases. People have been trained. Teams are experimenting with agents. There are pilots in marketing, HR, customer service, finance, IT. An AI community has been created. Someone has built an impressive use-case repository.
There is usually quite a lot happening.
And then somebody asks a very simple question:
What value have we actually created?
The room gets quieter.
This doesn't mean companies are losing faith in AI – quite the opposite. What emerges is something more subtle: a growing AI fatigue caused by the distance between the amount of activity taking place and the amount of business impact leaders can actually point to.
Adoption is high. Value at scale is rare.
The numbers make that tension difficult to ignore.
According to Stanford's 2026 AI Index, 88% of surveyed organizations were using AI in at least one business function in 2025, up from 78% the year before. Generative AI alone was being used in at least one business function by 70% of organizations. AI is no longer sitting at the edge of the enterprise – it is becoming part of everyday work.
At the same time, research paints a much less comfortable picture of value realization. Only 5% of companies in its research were generating AI value at scale, while 60% reported no material gains despite significant investment. What's particularly striking is that this hasn't reduced enthusiasm: 82% of CEOs surveyed were more optimistic about AI's ROI than they had been a year earlier.
Consider that combination for a moment:
- 88% adoption
- 5% value at scale
- and growing optimism
That does not look like a technology that has failed. It looks like a technology whose adoption has moved considerably faster than the organizations trying to absorb it – and that distinction matters.
AI clearly delivers – at the individual leven
There is another body of evidence that should not be ignored.
AI can already make people materially better at certain kinds of work. A large field study by Erik Brynjolfsson, Danielle Li and Lindsey Raymond followed 5,172 customer-service agents and found that access to a generative AI assistant increased issues resolved per hour by 15% on average. Less-experienced and lower-skilled workers benefited particularly strongly.
So when the increasingly common conclusion surfaces that "maybe AI isn't really delivering," the evidence tells a different story.
The technology is clearly capable of delivering. The more interesting question is why so much of that capability seems to disappear somewhere between an employee's desk and the company's P&L.
AI profileration is not AI transformation
Part of the problem lies in what has been measured over the past few years: AI activity.
- How many licenses?
- How many active users?
- How many prompts?
- How many use cases?
- How many pilots?
- How many hours of AI training?
All useful indicators when getting started. But none of them reveals whether the organization is fundamentally better at doing anything.
A company can have 200 AI use cases and still serve customers in essentially the same way. It can give every employee Copilot and still have the same meetings, the same approval layers, the same handovers and the same decision-making processes. It can use AI to write reports faster while continuing to produce reports nobody really needs. It can automate emails that probably shouldn't have existed in the first place. And it can save employees thousands of hours without ever deciding what those thousands of hours are now supposed to become.
That is where some of the fatigue comes from: AI proliferation has been confused with AI transformation. The two are not the same.
There is a human dimension to this as well
Employees have spent the last few years being told that AI is transformational. They have learned new tools, changed habits, attended trainings, experimented, read prompt guides, and tried assistants that worked brilliantly one day and strangely the next.
Managers have been asked to find use cases. Leadership teams have approved investments. Boards have heard repeatedly that AI will change everything.
Eventually, people quite reasonably want to know what exactly has changed. That does not make them resistant to AI – it makes them serious about it.
From "we are learning" to "we are changing"
What this points to is a healthier stage of the conversation taking shape.
The excitement of 2023 and 2024 made experimentation possible – that period was probably necessary. Organizations had to play with the technology before they could understand it. But experimentation cannot be the operating model forever. At some point, "we are learning" has to become "we are changing."
This is where the picture becomes clearer: the technology itself remains extremely promising. What deserves much more scrutiny is the way many organizations are implementing it.
Because if a fundamentally new capability is simply distributed across an organization designed for a different technological era, the results should not come as a surprise if they turn out to be incremental. The engine is not to blame if it is installed in the old car and nothing else is changed.
The companies getting serious value from AI seem increasingly to be doing something different. They are moving beyond isolated assistants and individual productivity gains and putting AI deeper into core operations. They are questioning processes, decisions, responsibilities and eventually the operating model itself. Meanwhile, research explicitly contrasts companies that graft AI onto existing processes with those that are redesigning processes from the ground up.
That is where the conversation needs to go next. Not away from AI – deeper into it.
The question leaders should really be asking
AI fatigue does not appear to be fatigue with AI itself. It looks more like fatigue with activity that does not lead to enough change.
And perhaps the most important question for leaders now isn't:
How much AI are we using?
It is:
What is our organization genuinely able to do better because of it?
And if that answer is still difficult to give, there is another question worth asking:
If AI is demonstrably making individual tasks faster, why is turning those gains into financial return proving so difficult?
That is where the next article in this series picks up.
Author’s note: The observations in this article draw on recurring patterns from our work with organizations navigating AI transformation, combined with publicly available research. Client situations have been generalized and anonymized; no individual organization or engagement is being described.
Sources: Stanford Institute for Human-Centered AI, AI Index Report 2026; Brynjolfsson, Li & Raymond, Generative AI at Work, Quarterly Journal of Economics (2025); Dillon, Jaffe, Immorlica & Stanton, NBER Working Paper on generative AI use across 66 firms and 7,000+ knowledge workers (2025); Stanford Digital Economy Lab, Enterprise AI Playbook (2026).
