CEOs, GenAI and the data underneath

Before asking what generative AI can do for a company, I would ask where its answers will come from.

The Information's discussion of CEOs, generative AI and data raises a practical leadership question: how much does a CEO need to understand before putting these systems to work?

Enough to ask useful questions. Which decision are we trying to improve? What information supports it? Who can see that information, and who is responsible when the answer is wrong?

Those questions quickly become specific. A sales team might want to know which customers are buying less. The answer depends on what counts as a customer, whether returns are included, which period is being compared, and how recently the records were updated. A fluent paragraph can conceal disagreements about every one of those things.

I would start with one application whose output someone can check. Give the team access to relevant data, define what a useful answer looks like, and examine failures alongside successes. That provides evidence for deciding whether to expand.

Data access also requires an owner. A system that can combine information across departments may expose details that an individual employee could not previously retrieve. Permissions need to survive that combination. So does the ability to trace an answer back to its source.

Review needs a place in the workflow. Someone should know when to verify an output, how to correct it, and when to stop using the system. Asking people to “keep a human in the loop” leaves too much undefined.

A CEO can delegate the implementation. Choosing the problem, setting the standard of evidence and assigning responsibility still require leadership. I would rather see one useful application with those decisions made clearly than a list of experiments nobody can evaluate.