The monthly aggregation includes all seven days of each week, but if the month starts or ends mid-week, the first and last weeks have different lengths. This creates an edge effect where month-boundaries can show artificial bumps just from the different number of days. Should we aggregate to calendar weeks instead?
Is this in absolute terms, or raw?
Why does this peak so consistently on Tuesdays? Is there a weekly pattern we should account for?
What's the minimum sample size? See /u/plg_theoden_damodred/p/plot-0064.
Missing Mexico data after 2012
The decomposition by customer tier is the right move here — it lets us see whether we're dealing with a universal effect or something tier-specific that might warrant different treatments.
That matches what I had.
That's fair.