A few years ago, “do you use AI?” was a box on a questionnaire. Now it opens a conversation, and the people asking have learned what a weak answer sounds like. Here are the questions we hear most often, and what separates a good answer from a policy document.
1. Where does AI sit in your process, and what can it do without a person?
A good answer is a map, not an adjective. It names each decision the AI proposes, each one it takes, and the limits that apply to both.
2. How was it validated before you relied on it?
A good answer shows results on data the system never saw, and says how many variations were tried before the one in production was chosen. Without that count, a good backtest says little.
3. How do you know what it read is real?
A good answer traces every fact behind a decision to the sentence it came from, and checks that link with ordinary code rather than another model.
4. What stops it from doing something it should not?
A good answer puts the limits outside the model, where it cannot talk its way past them, and shows a record of attempts to break them.
5. Can you show me why a specific decision was made?
A good answer is a replay, not a reconstruction: the inputs, the views, the checks that passed or held, and the person who signed, in order.
6. What does it cost, and who watches the cost?
A good answer meters the spend per decision and reports returns net of it.
7. What happens when it fails?
A good answer names the kill switch, the safe path that is always open, and the people who keep the decisions machines should not make.
The pattern
Every good answer is evidence rather than intent. A policy says what should happen; a record shows what did. Allocators have started to tell the difference.
If you want to rehearse these questions on your own process, that is a good first conversation.