Why Most Owners Get Nothing From AI

Part of the AI for the People Who Run Things series. Start with AI Won't Run Your Business.

In 2025, researchers at MIT went looking for the return on the AI gold rush. Companies had poured somewhere between 30 and 40 billion dollars into generative AI. The team tracked 300 public deployments, interviewed 150 leaders, surveyed 350 employees. Then they published the number that made the room go quiet. 95% of those companies saw no measurable impact on profit. None.

Five percent got real results. The other ninety-five got a demo and a bill.

Here's what makes that number useful instead of just grim. MIT was specific about the cause. It wasn't the models. The technology worked fine. The companies that failed didn't fail because the AI wasn't smart enough. They failed at the part that has nothing to do with AI: choosing what to point it at.

The "Use AI" Trap

Walk into most companies and the AI plan is, roughly, "use AI." Roll it out. Buy everyone a license. Run ten pilots and see what sticks. It feels like progress. It's the exact thing MIT watched fail 95% of the time.

When you deploy AI everywhere at once, you dilute everything. No single use gets enough attention to actually work. Each pilot stays shallow, nobody owns the outcome, and when money gets tight they all get cut together, because not one of them has a number attached to it.

I see the small-business version of this constantly. An owner buys the team a ChatGPT subscription, sends a message that says "start using AI to save time," and feels like the thing got done. Three months later nothing has changed. Because "use AI to save time" isn't a task. It's a wish. Nobody knew which of their actual jobs it was supposed to touch, so it touched none of them.

Gartner predicted that at least 30% of generative AI projects would be abandoned after the proof-of-concept stage by the end of 2025. The top reason wasn't technical. It was "unclear business value." Translation: nobody could say what the thing was actually for.

A demo answers "can it." It never answers "should we, here, for this specific job." Most AI projects die in that gap.

Why a Six-Person Shop Can Beat a Fortune 500

And here's the genuinely good news for a small owner. The thing that kills AI at big companies is the thing you don't have.

You don't have ten departments each running a vanity pilot. You don't have a committee that bought a platform nobody asked for. You have one operation, you can see all of it, and you already know which task is bleeding you. The winning move, pick one workflow, give it a number, deploy it there and nowhere else, is nearly impossible to coordinate across 6,000 people. It's a Tuesday afternoon for a shop of six.

You can decide it at lunch, try it that afternoon, and know by the end of the month whether it earned its keep. No procurement cycle. No change-management deck. No quarterly steering review where four VPs water it down. That speed is a real advantage, and almost nobody frames it to small owners that way.

MIT found the companies that did win mostly bought focused tools from specialized vendors and partnered closely. Those succeeded around 67% of the time. The ones that tried to build sprawling internal AI for everything succeeded about a third as often. Narrow and bought beat broad and built.

It's Not That AI Doesn't Work

Don't read the 95% as "AI is hype." Five percent of those same companies got rapid, measurable acceleration using the exact models everyone else had access to. The gap between the 5% and the 95% wasn't budget. It wasn't talent. It wasn't some secret model.

It was specificity. The winners aimed at one workflow and could tell you the before and after. The losers aimed at "the future" and couldn't tell you anything. And specificity is entirely in your control, no matter how small you are.

So don't be the ninety-five. Don't "adopt AI." Pick the single workflow where you can already feel the cost. The follow-up that's too slow. The quotes that take three days. The inbox nobody gets to until Friday. Put a number on it. Point AI at that one thing, watch it for two weeks, and leave the rest of your business alone until that one works.

The owners who get nothing from AI are almost always the ones who tried to get everything at once. The ones who get something picked a single job and stayed honest about whether it actually moved.

Part of the AI for the People Who Run Things series. Continue with The Jagged Frontier of Your Business.

---

This article reflects the state of AI tooling as of June 2026. What the models can and can't do reliably is still moving.

Sources

• Fortune: "MIT report: 95% of generative AI pilots at companies are failing" (2025), on MIT NANDA "The GenAI Divide: State of AI in Business 2025" (opens in new tab)
• MIT NANDA: "The GenAI Divide: State of AI in Business 2025" (report PDF) (opens in new tab)
• Gartner: "Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025" (July 2024) (opens in new tab)