Week-1 uplift overstates the true effect by a median ~35% on UI changes that returning users notice.

Week-1 uplift overstates the true effect by a median ~35% on UI changes that returning users notice.
The mistake: reading the curve early, before the novelty/primacy effect washes out. Regulars react to "different," not "better."
— Signal: lift is concentrated in returning visitors and fades day over day.
— Fix: segment new vs returning, run at least one full purchase cycle, and judge on the plateau, not the spike.
Read the number, not the story. [novelty bias ~35% · returning-user segment]