Most AI projects fail because nobody actually built anything.

We have a running joke internally. Whenever a prospect tells us they're six months into "AI transformation" with one of the big firms, we ask one question: "What's in production?" The answer is almost always silence. Then a slide deck.

This isn't an indictment of the people doing that work. Most of them are smart. The model is broken — not the staff. The model is: charge for thinking, not for shipping.

The shape of the failure

It looks like this every single time.

• 01Phase one — discovery. Two months. A document.
• 02Phase two — opportunity matrix. Workshops. Another document.
• 03Phase three — pilot scoping. A SOW for a pilot. Not a pilot.
• 04Phase four — pilot. Maybe. Half-built. Hands off mid-build.
• 05Phase five — "next phase" upsell. Restart the cycle.

At no point does anything that wasn't there before become there. Six months and the operations leader is still doing the thing manually.

Why this keeps happening

Three reasons, in order of size.

One — the firms are staffed for advisory work. They don't have engineers who can put a system in production. They have analysts who can describe the system that should exist. Different skill, different person, different cost structure.

Two — billing models punish shipping. If the firm is paid by the hour or by the month, finishing the work ends the money. The incentive is to refine the strategy document for another quarter.

Three — buyers buy the wrong thing. "We need an AI strategy" is a category error. You don't need a strategy. You need the report Linda spends three hours on every Friday to take twelve minutes. Strategy is a way of avoiding having to be specific.

“You don't need an AI strategy. You need the report Linda spends three hours on every Friday to take twelve minutes.”

What we do instead

We start with one process. We scope it in 48 hours. We quote it fixed. We build it in two to four weeks. We hand it over and we leave. Then we don't talk to you for six months unless you want us to. The system runs without us, because that was always the point.

No phases. No discovery theatre. No second round you didn't ask for. The deliverable is the working thing. If it's not working when we say it is, we're still there until it does — at our cost.

The honest test

If you're currently in a "transformation" program, ask the team this question: "What system is in production today that wasn't in production three months ago?" If the answer is a slide deck, a steering committee, or another phase — you have your answer.

You can keep paying for the strategy. Or you can pay someone to build the thing. Those are the two options. They are not the same thing.