The productivity numbers don’t lie. Or do they?
Most companies have now rolled out AI tools enterprise-wide. Licenses have been purchased. Trainings have been scheduled. Slack channels have been flooded with prompts. And yet, when leadership asks about the ROI, the room goes quiet.
This is not a new story. In 1987, economist Robert Solow looked at the data after years of massive corporate investment in personal computers and found something baffling: zero statistically significant improvement in productivity. Companies had bought the technology. They just had not changed how they worked. This became known as the productivity paradox, and it is playing out again right now with AI.
Here is the uncomfortable truth: most organizations are not suffering from a technology problem. They are suffering from a thinking problem. They got the tool. They skipped the strategy. I’m an AI transformation strategist, keynote speaker, and author of How to Do More with Less Using AI. I saw how AI changed my own team at Alibaba in 2018 and now I’m seeing the same mistakes happen in the wider industry.
Here are three signs your company is using AI wrong right now, and what to do instead.
1. You are measuring adoption, not outcomes
I was keynoting at a large Fortune 500 company the other day, and I heard that the big exec at the company was using adoption numbers by the number of people that logged into the tool. Yikes! I couldn’t believe that we were still looking at that as a verifiable number when it comes to AI adoption.
If your AI success metrics look like “percentage of employees who have logged in” or “number of prompts submitted per week,” you are measuring the wrong thing entirely.
Activity is not progress. A team that runs two hundred AI prompts a day but still produces the same output as before has not adopted AI. It has dressed up the same process in a new costume.
The organizations that are actually moving the needle are asking different questions: Has our decision-making speed improved? Have we eliminated work that used to create bottlenecks? Are we producing things that would have been impossible six months ago?
If you cannot answer yes to at least one of those, your AI adoption is theater.
The fix is straightforward, even if the work is not. Pick one workflow. Map what it looks like before AI. Map what it should look like after. Then close the gap. Do not measure how many people are using the tool. Measure whether the workflow is actually faster, better, or cheaper than it was before.
2. You are automating tasks without redesigning the role
History has a useful warning here. When the electric motor was invented in the 1880s, factory owners made a predictable mistake: they ripped out the giant steam engine and replaced it with one giant electric motor. They kept the same drive shafts, the same belt systems, the same cramped multi-story layouts. The factory was not faster. It was just quieter.
It was not until a new generation of managers realized they could put a small motor on each individual machine, and then completely redesign the factory floor around the actual workflow, that productivity finally exploded. That redesign took over thirty years. The technology alone was never enough.
Most companies are making the exact same mistake with AI right now.
A manager whose job was synthesizing weekly status updates and building PowerPoint decks now has AI that can do both in minutes. But no one told that manager what their new job is. So they spend the same time double-checking the AI’s work, tweaking a bullet point here and there, and calling it a productivity win.
Real AI adoption requires role redesign. Not just task removal. When you introduce AI into a workflow, the first question should not be “what can AI do?” It should be “what should this person focus on now that AI handles the rest?”
The answer to that question is where the actual value lives. For most knowledge workers, the answer involves more judgment, more creative problem-solving, and more direct ownership of outcomes. Those are not things AI can do for you. They are the things that become more valuable the more AI handles everything else.
3. You are outsourcing thinking before thinking
This is the quietest and most dangerous sign of all.
It happens when people stop forming their own view before going to AI. Instead of thinking through a problem, developing a hypothesis, and then using AI to pressure-test or expand on it, they open the chatbot first and adopt whatever comes back.
This is not laziness. It is a natural response to time pressure. But the long-term cost is steep. Judgment atrophies. People lose the ability to form independent views quickly. And when the AI is wrong, no one catches it because no one was thinking hard enough to notice.
I have watched this happen at large enterprises that were among the earliest adopters of generative AI. The first-year productivity gains were real. The second-year results were puzzling: output was up, but quality had flattened. When we dug in, the pattern was consistent. People had stopped arguing with each other, stopped stress-testing ideas, stopped pushing back. Because why bother when the AI already had an answer?
The best AI practitioners share a common habit: they think before they prompt. They arrive at the AI with a point of view, use it to challenge and refine that view, and leave with something better than what either they or the AI could have produced alone.
That is the collaboration model that works. Not AI as oracle. AI as sparring partner.
The bottom line
Solow’s paradox resolved eventually. Productivity did explode, but only after organizations stopped using computers to type old memos faster and started genuinely reinventing how they worked. The same resolution is available to companies today with AI.
But it requires changing how you work, not just what tools you use. Asking uncomfortable questions about which roles still make sense. Redesigning workflows instead of layering AI on top of old ones. And keeping the human thinking sharp, even when the AI could do it for you.
That is the only AI strategy that actually works. Everything else is just a more expensive version of the same old factory floor.
