For most of their history, industrial robot arms belonged to large manufacturers, and the reason was not only the price of the arm. A traditional industrial robot moves fast and powerfully and is not aware of its surroundings, so it has to be kept behind a physical barrier, which needs space and installation. It has to be programmed by someone who knows how, which means either an employee with that skill or an integrator's invoice. And because setting it up for a task is expensive, the task has to run for a long time to be worth it. Add those together and the minimum viable production run was large - far beyond what a workshop making dozens of things rather than tens of thousands could ever justify.
Collaborative arms changed each part of that arithmetic, and it is worth separating which change did the most. They are slower and weaker, which sounds like a disadvantage and is in fact the enabling feature: being slow and weak enough to stop safely on contact means they can work next to a person without a cage, which removes the space, the barrier and much of the installation. They are cheaper, though not cheap. And crucially, many can be taught by being physically guided through the motion rather than programmed in code, which moves setting up a task from a specialist skill to something a workshop's own people can do in an afternoon.
That last change is what actually matters for small runs, and it is consistently underplayed. When setting up a new task takes an afternoon rather than a fortnight and a consultant, the run length needed to justify it collapses. A workshop can put the arm on a job of a few hundred pieces, take it off, and put it on something else next week. The machine becomes a tool that gets picked up for a task, like any other tool, rather than a fixed installation that a production line is designed around. That is a different category of purchase, and it is the one that a small business can actually reason about.
The honest caveats are worth stating, because the marketing tends to skip them. These arms are good at repetitive, well-defined motions with parts that arrive in a predictable position and orientation - and getting parts to arrive predictably is frequently a bigger job than the robot itself, a cost that surfaces only after the arm is bought. They are poor at anything requiring judgement about a part that varies. They break even over a long period rather than immediately. The realistic candidate is a specific repetitive operation that a person currently does for hours, that is physically wearing or dull, and where the parts already arrive in a consistent way. That is a narrower target than the brochures imply, and it is a real one.