AI Capability

Why the word capability and not tool

A tool is something you own. A capability is something you are. The distinction matters because the failure mode of AI in mid-sized companies is almost never the technology, it is that the knowledge sat with one supplier or one enthusiastic employee, and left when they did.

An AI Capability is not something you buy or hire. It is something you become.

What it is made of

Four things, and all four have to be present:

• People who can operate and judge the system, not just use it
• Decisions that are actually owned, so someone can say yes or no when the model is wrong
• Data that means the same thing across departments, which is usually the hardest part
• Working methods that are written down, so the practice survives a resignation

Remove any one of these and you have a pilot, not a capability.

How you can tell you have it

A practical test: if the external party disappeared tomorrow, would the thing keep running and would anyone be able to change it? If the answer is no, you bought a tool. If the answer is yes, you built a capability.

Where it starts

Capability is built in a sequence, not in one go. First a diagnosis of where you stand, then one working use case built with your own people, then the working method spread across teams. Skipping the diagnosis is the common shortcut, and it is why so many organisations end up with an impressive demo that nobody uses.