Il Sole 24 Ore reports that 74% of Italian companies are in advanced stages of AI adoption, but only 21% of large companies have structured guidelines for AI use at the board level. The gap is exactly there: you buy, you experiment, you present to the board, but the people sitting at the table don't have a grid to evaluate what management brings them.
A board member who receives an AI PoC needs three things, not a white paper. First: which data the model runs on, and who has seen that data leave the company perimeter. Second: how you measure failure, not just success. Third: what happens when it gets things wrong in production, and who pays the bill.
Without those three questions, the board approves a demo. With those three questions, it evaluates a system. The difference between the 74% who adopt and the 21% who govern is all here: the people who wrote the guidelines stopped looking at slides and started asking for logs.
Why this matters for anyone building enterprise AI: if your board doesn't know what to ask, your PoC gets approved for the wrong reasons and fails for the right ones.