Why most companies get nothing from AI (and the commitment that changes it)

Every few weeks I talk with an owner who tried AI and got nothing. The story is nearly always the same story.

The company bought some subscriptions. A few people experimented for a while. Someone made a demo that impressed a meeting. Then the experiments faded, the subscriptions kept billing, and the quiet conclusion settled in: this technology is not for businesses like ours.

The conclusion is wrong. The test was.

The number behind the disappointment

MIT's NANDA initiative studied enterprise AI pilots and reported that 95% showed no measurable impact on profit and loss.

Treat the number with care. It comes from one report, it was not peer-reviewed, and it drew fair criticism. But the pattern it describes matches what I watch happen in company after company, and one more finding from the same research tells you where the problem lives: workers in over 90% of the firms studied were quietly getting value from personal AI tools while the official pilot stalled.

Read those together. Individuals figure this technology out. Institutions run tests that fail. The difference is not the tool, because it is the same tool.

The shallow test

Here is the pattern behind most of the failures I see, and I call it the shallow test.

A company buys access. Nobody's job changes. Nobody with authority to build things lives inside the technology daily. The work of teaching the tools what the business knows never happens, so every output comes back generic, and generic output convinces everyone the technology is generic.

The test was designed, without anyone deciding it, to produce a shrug. A shallow test can only produce test results.

I hold myself to the same standard I am describing. The difference in my results did not come from talent with the tools. It came from the decision to live in them with real work at stake, every day, until I understood what depth requires.

What the commitment looks like

The commitment that changes the outcome is not a bigger budget. It is depth, and depth has a specific shape.

Someone with authority to build has to own the technology as their daily work. The business's knowledge has to be captured and put into the system, because an AI that knows nothing about your company can only give you the generic layer everyone else gets. And the output has to be refined against a real standard until it holds, which takes judgment and repetition, the way any craft does.

In my practice, that shape is literal: hours of recorded interviews, a suite of brand guides built from them, and a Claude Fable project refined until the output is ideal for one specific brand. (The full architecture is in how one person and a trained system replace a marketing department.) But the principle is bigger than my method. Depth is what the 5% have in common.

What the commitment will not buy

Now the honest part, because you deserve the whole picture.

Commitment does not guarantee results, and anyone who promises otherwise is selling. A deep implementation still fails when the captured knowledge is thin, when the owner withholds the hours, or when the business problem was never one AI could touch. The technology multiplies what you feed it, including nothing.

And no implementation, shallow or deep, creates demand out of air. AI can build your presence, your content, and your consistency. Your reputation and your work still have to earn the attention those assets convert.

The harder question

The owners who conclude AI has nothing for them are usually reasoning from honest evidence. The experiments were run. The output was mediocre. The conclusion followed.

But the evidence came from a shallow test, and a shallow test cannot tell you what the technology is worth. It can only tell you what a subscription without commitment is worth, and that answer was always going to be nothing.

So before you close the question, ask a harder version of it. Was your test ever deep enough to tell?