The easiest AI project to greenlight is usually the wrong one. A chatbot on your website, a summarizer for internal memos, an assistant that drafts emails — these are easy to demo in a board meeting and easy to forget about six months later, because they don't touch the number anyone is actually accountable for.
The better filter is simpler than it sounds: where does this business lose money or leave money on the table because a human has to do something slow, repetitive, and judgment-light? Comp analysis that takes 80 hours instead of one. A research process that runs on founder pattern-matching instead of a repeatable system. A category screen that takes a quarter instead of an afternoon.
Those are the projects worth building first, because the return shows up in hours saved, deals closed faster, or revenue that was sitting on the table the whole time. Impressive demos are optional. Moving the number is not.
