Before you clip the outliers

A tail worth looking at rather than removing. revenue_usd on ab-test-checkout tops out at 396.58 against a median of 0.00.

The max is 2.8× the 99th percentile and 14.5 standard deviations above the mean. The median is 0.00, so the top of this column lives in a different regime from the middle of it.

select *
from ab_test_checkout
order by revenue_usd desc
limit 20;

Run that before deciding anything. There are three things an extreme value can be, and they need opposite treatment: a data error, a real but rare event, or a different population that got mixed into the same table. Only the first is safe to drop, and you cannot tell them apart from the summary statistics — you have to look at the rows.

A ratio this large usually means the third case. Something in this column is not the same kind of thing as the rest, and the honest fix is a filter with a stated reason rather than a percentile clip.

Clipping at the 99th percentile is a decision to discard your most interesting rows. Sometimes right. Never automatic, and never silent — if a number in a report was computed on clipped data, the report has to say so.