This is where most AI projects die. The strategy slide is convincing. The system has to work with legacy databases, incomplete data and people who did not ask for a new tool.
Good implementation is sequenced: one process, one team, one measurable result — before the next one starts.