Boards are being asked to approve AI investment against proposals written in the language of possibility. The technology is described, the potential is described, and the commitment is left implied. That is not dishonesty, it is what happens when a subject arrives faster than the vocabulary for governing it.
These six questions cut through it without requiring anyone in the room to understand a model.
One. What specifically gets cheaper, faster or better?
Named process, current state, expected state. If the answer describes a capability rather than a process, it is not yet a plan.
Two. Who stops doing what?
If nobody stops doing anything, the initiative has added work and the saving is theoretical. This is the same subtraction problem that sinks transformation programmes, arriving in new clothes.
Three. What is the number in twelve months, and who owns it?
A named person, not a function. AI initiatives are unusually prone to collective ownership, which is another way of saying nobody will be asked about it next year.
If nobody stops doing anything, the saving is theoretical.
Four. What happens when it is wrong?
Not whether it will be wrong. It will be. Who notices, how fast, what does the customer experience in the meantime, and what is the reversal. A plan without an answer here is a plan that has only imagined success.
Five. What data are we giving away?
What leaves the building, under what terms, and whether it can be used to train something you do not control. This is a commercial question about your own asset, not only a compliance question.
Six. What are we choosing not to do to fund this?
Every AI budget displaces something. Making the displacement explicit turns an enthusiasm into a decision, and it is the question that most reliably improves the proposal.
What good answers sound like
Plain, specific and slightly boring. Two or three named processes, a named owner, a number, an honest failure mode. Proposals that answer all six tend to be smaller than the ones that answer none, and they tend to work.
If a proposal cannot survive these questions, the right response is usually not rejection. It is to send it back for a fortnight, because the thinking is normally recoverable.