Why AI programs fail for unfashionable reasons

Why AI programs fail for unfashionable reasons

Every quarter a new tool arrives with a demo that works in the room and dies on the floor. The post-mortem is almost never 'the model was insufficient.' It is: nobody owned it, the data could not support it, and the people who would have to change their week were not in the meeting where it was bought.

That pattern is unfashionable because it does not flatter anyone. Vendors cannot invoice for it. Integrators cannot productize it. Internal champions cannot put it on a slide. So organizations keep funding the fashionable scarcity — more models, more licenses, more 'AI fluency' lunch-and-learns — while the unfashionable ones compound.

Ownership is the first. If the initiative does not have a named executive, a number, and a date, it is a hobby. Committees do not ship. Neither do steering groups that meet to admire a roadmap. Someone has to be able to say 'this is late, and it is my fault.' If that sentence is impossible, stop. An audit that surfaces this is worth more than a platform that conceals it.

Data is the second. Production AI does not run on the warehouse slide. It runs on whether identity, lineage, and access actually exist for the workflow you claim to automate. A copilot on top of three unsynchronized systems is a faster way to be confidently wrong. The expensive work is unglamorous: which table, which owner, which exception path.

Change is the third. Technology programs that do not alter how Thursday looks for a named role are decorations. Training is not a slide deck. It is time on the calendar, permission to work differently, and a manager who will not punish the new habit. If the last ERP or RPA program did not stick, the next model will not either — unless you treat change as a dimension with a score, not a communications plan.

AIDA exists because these three, plus strategy, stack, and governance, are the actual failure modes. A 5-minute score will not replace an evidentiary audit. It will tell you which of the six is the constraint so you stop spending on the other five. That is the unfashionable move. It is also the one a CFO can defend.