S&P 500 companies mentioned AI in earnings calls at record rates between 2024 and 2025. Almost all of them described implementation as entirely positive. The productivity data tells a different story.
Most macro studies of productivity growth find limited evidence of a significant AI effect. Even firms that say it is useful find little evidence of transformative gains. The investment is real. The returns are not showing up.
There are a lot of theories about why. Most of them focus on the technology. They suggest it was the wrong tools, wrong integration, too early in the adoption curve. Some theories focus on governance and some others on training.
I think the diagnosis is simpler, and more uncomfortable. Most organisations have not treated AI as something that requires deliberate deployment. They have treated it as something you announce.
What “use AI” actually means in practice
In most organisations right now, AI implementation looks roughly like this: a directive comes from leadership, tools get licensed, and people are told to find ways to use them. The expectation is that productivity will follow.
ManpowerGroup’s 2026 Global Talent Barometer, drawing on nearly 14,000 workers across 19 countries, found that while regular AI usage jumped 13% in 2025, confidence in the technology plummeted 18%. “Workers are being handed tools without training, context, or support,” the researchers noted.
This is not a technology problem. It is a deployment problem. And it is entirely predictable when you look at the conditions people are being asked to operate in.
Workers are being told to use tools they have not been trained on, to find applications they have not been helped to identify, against a media backdrop telling them that if AI succeeds, it may cost them their jobs. 43% of workers fear automation may replace their job within the next two years. Asking people to enthusiastically implement a technology they associate with their own redundancy, without involving them in the plan, is not a change management strategy. It is a recipe for skepticism and even sabotage.
Deloitte’s TrustID research found that trust in company-provided AI tools fell 31% between May and July of 2025, a short window that suggests the drop was driven by lived experience of implementation. People tried the tools, weren’t able to use them well, and trust collapsed.
The productivity gap is not a mystery. It is a consequence.
What deliberate deployment actually requires
AI is not a thing you use. It is a tool you deploy, and deploying it well requires two things most organisations are skipping.
The first is understanding your own organisation. Before you can identify where AI will add value, you have to understand how work actually flows through your business: where decisions get made, where processes are slow, where people are spending time on low-value tasks that could be automated, and where the genuinely human work sits that should not be replaced but freed up for. This is not a technology audit. It is an operating model question. You cannot identify the right applications for AI without first understanding the system you are deploying it into.
The second is involving the people doing the work. This is not a soft point about feelings. It is a practical one. The people doing the work know how it actually operates, as distinct from how the process map says it operates. They are the best source of information about where the friction is, where the bottlenecks are, and where AI could genuinely make things easier. Talking to them before implementation, not after, produces better diagnostic information and a fundamentally different change dynamic. People who have been consulted, who can see that there is a plan, and who can see that the plan is designed to make their work better rather than to eliminate them, behave differently from people who have had a tool dropped on them and been told to get on with it.
This is also where the job displacement narrative does real damage. If the only story people hear about AI is that it takes jobs, and leadership is not actively telling a different and more specific story about what this implementation is for and what it means for the people in the room, the trust gap fills with fear. Fear is an effective blocker.
The organisations getting this right
The organisations I see making genuine progress share a few characteristics. They have done the diagnostic work. They know which problems they are trying to solve and why those problems matter. They have involved the teams closest to the work in identifying opportunities, which means those teams have both better ideas and a reason to want them to work. And they are telling a clear story. That story is not “AI will make you redundant” and not “AI will fix everything,” but “here is specifically what we are doing, here is what it means for you, and here is how your work changes.”
That combination, of deliberate deployment, genuine consultation, and honest communication, is not complicated. But it requires treating implementation as an organisational design challenge, not a technology rollout.
Most organisations are not failing at AI. They are failing at the step before it.