Before asking AI for answers, ask it for questions

We tend to think of generative AI as an answer machine. Give it a problem and ask for a solution. Upload a process and ask how to improve it. Describe a business problem and ask what should be automated. I increasingly think one of its more useful organisational roles is the opposite. Ask it to question you. Take something ordinary such as your purchasing process. Rather than asking AI how to automate it, describe how you believe it works and ask what it does not understand. Where does the process actually begin?

Who provides the information? What happens when information is missing? Who decides whether a purchase should proceed? Where does work wait? Does anybody enter the same information twice? What happens when the normal process does not fit? Where is judgement required? Some questions will be easy to answer. Others may expose things you have assumed but never checked. You may need to speak to somebody doing the work. You might need to examine some actual examples. Perhaps two people describe the same process differently. That is useful. The objective is not for AI to infer the correct answer. It is to help expose what the organisation does not yet know.

This changes the role we give the technology. AI is no longer pretending to be the expert who understands our organisation from a few uploaded documents. It becomes a tool for interrogating our own understanding. That distinction matters because a process map, procedure or manager’s description is only a representation of the work. A language model can produce an impressively sophisticated analysis of an inaccurate representation. So try giving your AI a different instruction: Do not recommend improvements yet. Help me understand what I would need to know before I could improve this work responsibly. Then investigate the questions it raises. Sometimes a better question is considerably more valuable than a plausible answer.