A simple test for whether an AI use case is worth building at all.

Ask three questions: is the task repeated, is the input consistent, can a human check the output fast. Two yeses, build it.

Myth: every workflow should be automated. Reality: some tasks are cheaper left to a person.

Do this: measure the task by hand first. Not this: automate a process you have never timed.

We almost automated a report nobody read. Timing it by hand killed the project, and that was the win.

The best AI decision is often deciding not to build it.