First of all, we need to consider that the samples we analyse are just that: samples that should represent something bigger, such as the process, the batch, etc. Proper sampling procedures need to be in place to make sure that sufficient unit samples are pulled from e.g. the process in an appropriate way that represents what we need to learn from the process.

Then the samples arrive in the lab and we need to analyse them. This is the part where the qualification/validation/method knowledge is one part of the answer. The other is regulatory requirements and the risks you are willing to take. 

Ultimately, the question is not only whether the analysis reliably describes what is in the sample, but also what we need to learn from the whole process. I can think of very little situations where one sample analysis would be sufficient to represent the whole process/batch/etc.

Please also note that when a sample is pulled and delivered to the lab for analysis, good practices dictate that the sample should not be further divided in sub-samples. If more replicate analyses are needed, multiple samples need to be pulled.