Plain Reckoner

Display sites get less traffic in December than in January, and earn a third more per visit

Across 99 display-only businesses, detrended so growth cannot leak into the calendar, December earns 1.34x January per pageview while receiving only 0.95x the pageviews. The Q4 lift is entirely an ad-rate effect, and the traffic curve works against it.

How it was measured
Every sold or listed business monetised by display advertising and nothing else, with twelve consecutive months of both revenue and pageviews covering all twelve calendar months. Each listing is detrended in logs against its own window before the seasonal index is taken, because businesses are listed on an upslope and that growth would otherwise land in December. The index is the median residual by calendar month; 1.00 is the listing own annual level. Reproducible with scripts/measure-rpm.mjs --seasonality.
Evidence current as of

Listen — read from this page, nothing added

Watch it plot 35s · narrated · the same figures as above
Revenue per thousand pageviews, indexed to each business own annual level, across 99 display-only businesses. Jan: 0.855 Apr: 0.965 Jul: 0.958 Sep: 1.054 Nov: 1.137 Dec: 1.142
Revenue per thousand pageviews, indexed to each business own annual level, across 99 display-only businesses.

Everyone in display advertising knows Q4 pays. What almost nobody separates is whether that is more people or more money per person — and the answer decides what you do about it.

The two curves move in opposite directions

Indexed against each business's own annual level, so 1.00 is its normal month:

MonthRevenue per pageviewPageviews
January0.8551.041
February0.9810.980
April0.9650.993
July0.9581.040
September1.0540.988
November1.1370.973
December1.1420.990

December is not a traffic month. These sites get about 5% FEWER pageviews in December than in January, and still earn 34% more per visit.

The whole Q4 effect is what advertisers pay. The audience curve is quietly working against it — which means the revenue swing you feel is smaller than the rate swing actually is.

What that changes

Publishing harder in December does not buy the premium. The premium is per impression, and you get it on the traffic you already have. Effort in December buys ordinary traffic at an extraordinary rate; the same effort in September buys traffic that will still be there when the rate arrives.

January is the real event, and it is worse than it looks. At 0.855 it is the deepest month of the year, and it arrives with the year's HIGHEST traffic. Every visitor is worth about a quarter less than in December. A site that plans cash around Q4 revenue meets that in six weeks.

The control, which limits the claim

Amazon-monetised businesses, computed the same way over 744 listings, show a December-to-January revenue-per-pageview ratio of 1.303 — against display's 1.336. Nearly identical.

So "display is unusually seasonal" is not supportable, and we are not claiming it. The Q4 rate lift is a feature of online commerce generally.

What IS specific to display is the shape: Amazon's December pageviews are flat against January (0.996) while display's fall to 0.951. For an Amazon business the two curves point the same way. For a display business they point apart, and only one of them is in your control.

Why the detrending matters

Businesses are listed for sale on an upslope — that is when people sell. Take a raw monthly average across a listing's window and that growth lands in whichever months sit at the end of it, which for a lot of listings is Q4.

You would then publish a seasonal effect that is really a selection effect. So each business is detrended in logs against its own twelve months first, and what is reported is the residual.

What would weaken this

Pageviews come from analytics-connected listings, so this is the verified subpopulation rather than all display sites. Revenue is seller-reported and broker-vetted, not an ad-network export. Detrending is linear in logs, so a business with a mid-window step change leaks a little into the index. And each business contributes one twelve-month window, so calendar years are unevenly represented.

Grade this B: the method is sound and reproducible and the control is real, but n=99 and the inputs are P&L rather than platform data.

Tracked as Q4-IS-RATES-NOT-TRAFFIC. If a number here is wrong, tell us and we will correct it in place and say what changed.

One email per finding

Get these as they land

One email per finding — the number, the method, and the date it was true. No schedule, because measurements do not arrive on one.

Data controller: Plain Reckoner. Contact [email protected].