A leadership team approves an AI tool to help shortlist candidates.

Six months later, someone who was rejected asks why they were not selected. The team looks at the system's output. They can explain who the tool ranked highly, but not why this particular candidate was screened out.

That is not a technology failure. It is a decision-making failure.

AI does not remove the need for a defensible rationale. It moves the point at which that rationale is needed. The awkward questions still land on the people who made the decision - including the decision to let a system influence it.

The problem is that AI adoption often starts with what the technology can do rather than what the organisation needs to be able to stand behind.

A team finds a way to automate shortlisting, sift applications or identify candidates who appear to meet the criteria. The system produces an answer. Everyone moves on.

Until someone asks: 'why?'

That question changes everything.

The organisation may be able to explain how the system was configured. It may be able to describe the model, the data or the criteria. But none of that necessarily explains why this particular candidate was treated this particular way.

And "the system said so" is not a rationale an organisation can own.

The answer is not to document everything or make every AI-assisted decision slow and bureaucratic. It is to decide beforehand what the organisation needs to be able to explain when a decision matters.

What would we say to the person affected? What evidence would we show? Could the person making the final decision understand and stand behind the reasoning?

If those questions cannot be answered without deferring to the model, the organisation has not decided what its own reasoning is.

That distinction matters because AI-assisted decisions are rarely experienced by the people who approved them. They are experienced by the person who does not get the interview, the candidate who is rejected, or the employee whose progression is affected.

They did not choose the system. They did not design the process. But they are the ones who have to live with its decision.

That makes the human judgement around the AI more important, not less.

The practical discipline is simple: before using AI in a decision that could affect someone, ask what happens when they ask why.

If a human can review the evidence, understand the output, exercise judgement and explain the decision, the AI can support that process.

If nobody can explain the decision without pointing back to the system, the organisation has handed over more than a task. It has handed over part of its responsibility.

Leaders do not need to understand every technical detail of every model their organisation uses. They do need to know where accountability sits.

The question is not simply, "What can this AI do?"

It is "What will we say when someone asks us why?"

If the answer is not clear, the organisation is not ready to let the system make that decision.