People add AI where it looks impressive. The better question is where it changes a decision you can actually measure. Three tests before you let any tool into your process: 1. Decision, not decoration. Does its output change what you buy, hold, or skip? If it only produces nice-looking summaries you never act on, it is decoration. Cut it. 2. Measurable before and after. Pick one number you already track — hours spent screening, holdings you misjudged, questions you left unanswered. If the tool cannot move a number you name in advance, you cannot tell whether it worked. 3. Checkable, not trusted. Never let output you cannot verify touch a real decision. If you cannot trace a claim back to a source and confirm it yourself, treat it as a rumor, not research. Flashy fails all three quietly. Useful passes all three on purpose. Where in your own process would AI actually change a decision you could measure?