There is a problem on website data, where new features are only touching a subset of customers and you need results for every single one.

You wont be partitioned for this case, but the compute you need is just filtering out this set.

But sampling wont get what you want especially if you are doing QC at the business team level about whether the CX is behaving as expected.