When is AI not worth it for a business?

When the problem is small, rare, or already solved by something off the shelf. If a forty-dollar-a-month tool does it, buy the tool, don't commission a build.

It's also not worth it when you'd be automating a process that's broken. You'll just get faster wrong answers. And it's not worth it when nobody owns the outcome: AI that helps you think only pays off if a human is making a better decision on the other end. I'd rather tell you it's not a fit than sell you a build you don't need. Useful beats billable.