Revenue per user, before and after trimming

Zalora tested a personalised recommendations rail. Revenue per user is a heavy-tailed metric: a single stylist buying a season's stock can move a variant's mean on its own, and which arm that buyer happens to land in is luck, not treatment effect. The usual defence is to trim the extreme tail before comparing.

From user_revenue (2,000 users), return each variant with:

- mean_revenue — the plain average, rounded to 4 decimal places - trimmed_mean — the average over only those users at or below that variant's own 99th percentile of revenue, rounded to 4 decimal places

Order by variant.