A large share of annual American retail activity happens in a short stretch between late November and the end of December. Every analytical method built on the rest of the year fails during it.
The baseline stops being a baseline
Forecasting and anomaly detection work by comparing current behavior to a recent period. In a season where volume multiplies within days, the recent period describes a different world.
Models trained on typical weeks flag normal holiday behavior as anomalous, and flag genuine problems as ordinary seasonal movement.
Which means the period requires its own comparison basis, usually the equivalent weeks of prior years, rather than any rolling window.
Shifting dates misalign the comparison
The holiday shopping period is anchored to a floating date, so the same calendar week contains different phases of the season from one year to the next.
Comparing week over week against last year therefore compares a peak against a build-up, producing swings that mean nothing.
Aligning by days relative to the holiday rather than by calendar date fixes it, and almost no reporting tool does this by default.
Testing during peak is unreliable
Audience composition changes completely, with gift buyers purchasing for others under time pressure. Their behavior does not represent the year-round customer.
A test that wins in December frequently fails in February, because it was optimized for a shopper who has left.
The countervailing pressure is that traffic is highest then, so tests reach significance fastest exactly when their results generalize least.
Attribution windows overlap themselves
Purchase cycles compress, and multiple campaigns touch the same shopper within days, so attribution windows overlap in ways they do not the rest of the year.
Every channel can credibly claim involvement, and the sum of channel-reported conversions exceeds actual orders by a wide margin.
Reading the season through platform-reported numbers therefore overstates everything, and only a total-revenue view keeps the arithmetic honest.
The following quarter is part of the season
January carries returns, gift card redemption and the demand pulled forward by holiday promotion, so it cannot be read as a normal month either.
Judging the season before that settles produces a flattering number, since the costs of the peak arrive after the revenue does.
Teams that define the season as running through the return window get a truer figure and a less pleasant one, which is generally the point of measuring.