Pooled datasets are created when the data from multiple related trials are "joined" into one larger dataset for analysis. Because the contributing trials often have different data specifications, pooling involves a lot of data manipulation to get common variable definitions.
It means that somebody needs a one-time report. It could be anything:
How many patients does doctor XYZ have?
If doctor XYZ dropped out of the trial, what's the age distribution of the patients who are left?
Who are the patients that had batch 12345 of the drug?
If this hurricane hits Florida, how many doctors and patients are affected?
These are not questions that you can program in advance, but a general knowledge of SAS procedures will help you satisfy these kinds of requests. I'm sure there are many other examples -- the thing is that these requests are not really predictable -- it could be anything -- that's why they're called "ad-hoc".