Which AI decisions deserve your time
Set one reference unit and call it X. It is close to what your company already approves without discussion. Every open AI decision then falls into one of three classes, and the handling is different in each.
Delegated
Decided by line management inside published guardrails. Most AI decisions belong here, and most of them currently get more attention than they are worth.
Decided by the leadership team
A written case, a named owner, stage gates, and a stop criterion. Two weeks of work, once, rather than eighteen months of steering meetings.
Decided by the board
At this size the commitment reshapes work that crosses several processes, which is why no single owner can decide it. The question stops being what to buy. In most companies this class is empty.
What is left after the sorting is usually one or two places where AI is worth a serious effort.
The point of the rule: it takes work away as well as adding it. That is what makes it survivable past the first quarter.