You are tracking daily completed rides on the Didi Chuxing rider app and see a decline right after a dispatch flow update. The series has strong weekday and holiday effects, so raw trend comparison is misleading.
How would you use time-series trends to separate seasonality from a real product issue?
Separate recurring seasonal patterns from underlying trendEstimate whether the post-change drop remains after controlling for seasonalityTest whether the intervention effect is statistically distinguishable from noiseTranslate the result into a product diagnosis