Why do AI projects fail? Because nobody wrote down what “working” looks like.

You walk out of the slummiest airport known to man. Someone sent a car. You don't know who, but a chauffeur is saying your name with the same enticing power of your mom talking to you in the womb for the first time. You don't know where you are going because you have never been there. All you know is, it must be going somewhere important. So you mindlessly get in, because there is a self-imposed FOMO cocktail being stirred inside you with a very shiny hype-induced urgency spoon.

Projects that did define success metrics upfront succeeded 54% of the time. Projects that didn't: 12%.

Industry consensus generally is "it's just an alignment problem." But alignment on success criteria requires that the criteria exist somewhere readily accessible. Also mostly because if it is not accessible, how do you consider its use case, or AI automation readiness?

Most of it doesn't. It lives undocumented in the judgment of the people doing the work, the minds of your most lethal strategists and the memory of your most talented analysts. This data is the blood in the veins of your organisation that nobody bothers to pulse check until it lands up in the ER and the specialist pulls up in a Bentley. Because emergency, and AI is going to bury this one.

You can't align people around criteria that don't exist anywhere except inside someone's head.

I deeply believe that leaders and decision-makers aren't failing to agree, they're being asked to agree on something nobody has ever written down, and they are flying completely blind. They are at the airport and that car is looking very, very comfortable right now.

The alignment gap is a symptom. Undocumented tacit knowledge is the disease that is going to cost you money, and then die anyway.

Berkeley's California Management Review made this point plainly this year: most companies treat AI as a data and technology challenge. The ones actually making progress have figured out the real differentiator isn't the data or the model. It's the tacit knowledge sitting in the judgment of their people, and whether that judgment has ever been made visible.

In a conversation this week with an operations and people strategy leader, all I could think about was this: AI cannot function without data, data means nothing without structure, and structure doesn't exist until every process, every SOP, every piece of human know-how has actually been documented. Not stored. Documented, and with the reasoning intact.

So before talking about AI, ask a harder question: "is our data ready?" Could your team write down, right now, what "this is working" actually means, in language specific enough that two people would agree on it?

If not, that's not a data problem. That's the actual project.

Where to start

If you want to see where your business stands, take the free AI Readiness Check. It takes three minutes and shows how much of your business lives in your head. If you already know which process hurts most, the Process Automation X-Ray gets it out of your head and scores it for automation.