For example, an RC filter will operate exactly as it should only if the source and load impedance are exactly what the filter has been calculated for. Presented with real-world source and load, the response will be somewhat different.
Let's say you want to design a filter, but you know that, for some reason, the actual load may vary; you have to choose a somewhat arbitrary design load. Let's say you choose a design load impedance of 600 ohms, at some time, you'll have to find out what happens when the load diverges from the design criterion, and then, you'll find out there's a point where it becomes so out of data that you deem it the max tolerable divergence (in fact there are two points: the upper limit and the lower limit). Eventually, you'll end up knowing that your filter is valid (with a reasonable tolerance) from maybe 400 ohms to 2kohms. If it doesn't satisfy you, you'll have to either change the target load or change your expectations.
All this is called sensitivity analysis: in theory, you should do it for every component, because of tolerances, you have to make sure you target will be met with every component drifting at the extreme of its tolerance range. In practice, you become more and more familiar with what is significant and what is irrelevant.
