18 January 2026 · Methods
When ARPU hides a price experiment
Average revenue per user is a ratio. If the denominator and the price vector both move, the ratio will tell a story that nobody ran.
A GB company tests a higher list in Scotland for six weeks. New logos arrive slower. Existing Scottish accounts stay on grandfathered prices. Company ARPU ticks up because the remaining new joiners in England bought annual plans. Nobody intended that sentence, yet it appears in the Monday deck.
Tag every invoice line with experiment_id, even when the id is “none.” Holdouts need the same field. When you plot ARPU, plot it inside an experiment_id, not across the company, until the test is declared over and grandfathering is documented.
If your billing tool lacks a custom field, keep a side table. The studio would rather you maintain a clumsy spreadsheet than a confident average.
ARPU also moves when you delete free users from the denominator. Say so. A cleaned ratio can be legitimate; an unexplained jump cannot.
We do not treat this article as a substitute for a proper experiment design. It is only the measurement hygiene we require before anyone claims the test “worked.”