Where pixel-level precision earns its keep
Medical imaging is one of the clearest cases, a radiologist or oncologist needs a tumor's exact boundary traced in a scan, not a rectangle around its general area, because treatment planning and progress tracking depend on precise size and shape measurements over time. Self-driving cars need pixel-precise understanding of the drivable surface, distinguishing road from curb from shoulder at the exact boundary, since a few misclassified pixels at the edge of the road can matter for safe path planning. Background removal tools, like the kind used for product photography or video call backgrounds, need a clean outline of the subject to cut away everything else, and any leftover box-shaped chunk of background would be immediately visible.
In each case, the task fundamentally requires knowing the exact shape, not just the rough location, of something in the image. That's the pattern to look for when deciding whether a project actually needs segmentation, ask whether the downstream use of the output cares about shape and exact boundaries, or whether knowing 'it's roughly here' would be just as useful.