We are asked to review a lot of AI strategies. The ones that succeed tend to answer three questions early and honestly; the ones that stall usually cannot answer any of them.
1. What decision or task changes?
Not “we will use AI in customer service” — which task, done by whom, changes how? If you cannot name the person whose day is different, there is no project yet.
2. Where is the data, and who owns it?
Almost every delay we see is a data access question wearing a technical costume. Establish ownership and permission before design, not during build.
3. What happens when it is wrong?
Every model is wrong sometimes. A plan that has not named the failure mode and the human check has not been thought through — and will not clear information governance.