Why Your Company's AI Rollout Isn't Working (And It Was Never the Tool)

The NBER surveyed close to 6,000 executives across the US, UK, Germany, and Australia earlier this year. Eighty-nine percent of them said AI had made no measurable difference to productivity at their company in the past three years. Not "some difference." None.

Then there is the MIT Media Lab report, the one that has been quoted everywhere: 95% of corporate generative AI pilots delivered zero return. Companies had poured tens of billions into this collectively. Almost none of it showed up on a P&L.

I see the smaller version of that number every time I run a session. I walk into a room where nearly everyone already has ChatGPT open on their laptop, and when I ask what has actually changed about how they do their job, the honest answer is usually nothing. They have the tool. They have had it for months. The work still gets done exactly the way it always did, just with a chatbot tab open in the corner.

If that is you, I want to stop you right there. You are not behind. You are in the majority. The tool is not the problem. Almost nobody's is.

What "no productivity impact" actually looks like

It rarely looks like failure. It looks like this.

A company buys the licenses. Everyone on the team gets access to ChatGPT, or Copilot gets switched on across the org. Someone sends a memo. Maybe there is a lunch-and-learn. And then, three months later, almost nothing about how the work actually gets done has changed.

People still do the same handoffs. The same Tuesday report gets built the same way, just with a chatbot tab open that mostly sits there unused. The tool exists. The habit does not.

That is not a training failure in the "nobody read the manual" sense. It is a sequencing failure. The rollout started with "here is the tool" and skipped the step where someone asks which part of this job is actually worth changing. I have written before about why the workflow has to come before the tool, and this is that same mistake showing up on a company-wide scale.

The MIT report says the same thing, from the other direction

The MIT researchers went looking for why pilots stall, and their answer was not "the models aren't good enough." It was friction. Companies chase the exciting, general-purpose use case, a chatbot for everyone, instead of the boring, specific one: automate this one recurring task that currently eats four hours a week from one team.

The 5% of pilots that did work shared a pattern. They were narrow. They were attached to a real, named workflow. Someone could point at the before and the after.

That is the whole difference between the 95% and the 5%. Not better prompting. Not a better model. A better starting question.

Why this matters more in the Gulf right now

Adoption here has been fast. Leadership teams across the UAE are telling staff to "use AI" the same way they told them to use Slack or Teams a few years ago, often from the top down and often quickly. But telling someone to use a tool is not the same as showing them which piece of their actual job it should touch first.

So the gap the NBER and MIT numbers describe is the same gap I see walking into a session here. Everyone has access. Almost nobody has a habit. If anything, the faster the mandate came down, the wider that gap tends to be, because the tool arrived before anyone mapped the work.

How to actually be in the 5%, not the 89%

Here is the version of this that works, and it is less exciting than "roll out AI across the company," which is exactly why most people skip it.

Pick one task, not a department. Something you personally do every week that quietly eats a chunk of your time. Not "customer service." One task: the weekly status report, the first-draft reply to a recurring type of email, the meeting notes nobody has time to write up properly.

Map it before you touch a tool. Write down the actual steps, in order, the way you really do it, not the way it is supposed to work on paper. Most people skip this because it feels slow. It is the fastest step in the whole process, because it is the only one that tells you where AI actually fits.

Ask where AI fits, not whether it fits. Once you can see the steps, the use case usually finds itself. Maybe it is drafting, maybe it is summarizing, maybe it is just formatting. It is rarely "replace the whole task."

Do it manually once with the AI alongside you, not instead of you. Run the task the normal way, but with the AI drafting in parallel. Compare. Keep what is actually faster or better. Throw out what is not.

Only then decide if it is worth repeating. If it saved you real time on a real task, do it again next week. That is the whole test. Not "does this feel impressive." Does it hold up a second time.

If you want to see what that looks like when it is done properly with a whole team, I wrote about what onsite AI training actually involves, and a more general starting guide on how to use AI at work without any technical background.

What this means if you're the one being told to "use AI more"

If you are feeling behind because the AI subscription your company bought hasn't transformed your week, the data says it clearly: it hasn't transformed almost anyone's week. Eighty-nine percent of executives are looking at the same flat line you are.

The fix is not a better tool, and it is not more hours spent exploring one. It is picking one real task, looking honestly at how it currently works, and only then bringing AI in. That is a smaller, slower-sounding first step than "roll out AI everywhere." It is also the step the 5% actually took.

That is the whole thing I do with teams, onsite, on their real work. If your rollout has gone quiet since the launch email, tell me what your team actually does all day, and we will find the first task worth changing.

Frequently asked questions

Why do most companies see no productivity gain from AI?

Because the AI gets added to a workflow that never actually changes. Access is not the same as adoption. Teams get the tool but no one identifies which specific task it should replace or improve first, so habits never move.

Is the AI productivity gap a Dubai or Middle East specific problem?

No. It shows up in every market the NBER study covered, including the US, UK, Germany, and Australia. In the UAE the pressure to adopt is often faster and more top-down, but the underlying gap between access and habit is the same everywhere.

What is the fastest way to find where AI actually helps in my job?

Pick one recurring task that eats real time each week, write down its actual steps, and only then look for where AI fits. The use case becomes obvious once the workflow is visible. Trying to spot it before that usually fails.