Most AI work in companies falls into three buckets. Productivity tools that speed up tasks, rebuilds of things you already do, and things that were not possible before AI. Each needs a different budget, a different timeline and a different definition of success, and companies routinely mix them up. That is how a crowded pipeline gets mistaken for transformation. I will go through what each bucket actually returns, what has to change for those returns to show up, and why most of the difficulty turns out to be organisational rather than technical. Field notes from real projects and deployments, not another transformation framework.