We know something is wrong but we do not know what to ask for

This is a better starting point than arriving with a detailed specification, and I mean that.

A client who has already decided the answer has usually decided it based on the most visible symptom. A client who says "something is wrong here" has left the diagnosis open.

How we find it

You do not need to arrive with a solution. You need to be able to describe a bad day.

What went wrong. Who had to fix it. How long it took, and how often it happens. That is enough to start, and from three or four of those the pattern is usually obvious to an outsider even when it is invisible from inside.

The questions that unlock it

• What is the thing you keep meaning to fix and never get to?
• Which report does someone build by hand, and who actually reads it?
• Where do you find out about problems from the customer rather than from your own systems?
• If you were off for a fortnight, what would break first?

That last one is unfairly effective. It finds the single points of failure immediately, and they are almost always a person rather than a system.

What I will tell you

Sometimes the answer at the end of this is that you do not need software. You need a decision nobody has made, or a process that should be deleted rather than automated.

I would rather say that than build something to paper over it. It costs me the build and it earns the next three.