news blog × Adverge · 22-day 50/50 A/B test
+49%
more revenue on the same traffic, and ahead on every single day.
A news blog split its live traffic 50/50 between Adverge and its incumbent setup for 22 consecutive days, across 1.04 million sessions and 3.4 million pageviews.
Revenue per 1,000 sessions ran at $1.09 against $0.73. Adverge finished ahead on all 22 days, by never less than 22%.
It did it by returning more impressions and selling them for more. Both, not one at the cost of the other.
Revenue · 22 days
+49.4%
on period totals, Aug 27 - Sep 17
Revenue per 1,000 sessions
$0.73 → $1.09
page RPM moved $0.23 → $0.34
Days ahead
22 of 22
narrowest day +22.3%, widest +94.5%
Twenty-two days, side by side.
Revenue per 1,000 sessions. Traffic was split 50/50, so each side is measured against half the site’s sessions that day.
Two things are happening at once, and the page is clearer if we say both. Adverge holds a flat $1.09 per 1,000 sessions from the first week to the last. The prior setup slides from $0.78 to $0.67 across the same window. So the margin widens over the test because the incumbent falls away, not because Adverge climbs.
Where the lift came from.
A revenue gain can come from selling more impressions or from selling them for more. Here it was both, and the two multiply out to the headline exactly.
Impressions
×1.30
Adverge returned 4,549 impressions per 1,000 sessions against 3,512: 2.36M against 1.82M across the test.
eCPM
×1.15
And it sold them for more, not less: $0.240 against $0.208 per thousand impressions.
Revenue
×1.49
The 22-day result. More inventory monetized, at a higher clearing price, on identical traffic.
This is the check worth running on any yield test. A stack can lift revenue by simply printing more impressions at a falling price: volume up, eCPM down, users worse off. That is not what happened here: both terms moved the same way.
22 days.
22 wins.
Each bar is one day’s difference in revenue per 1,000 sessions. Not a monthly average hiding bad weeks. The narrowest day still came in 22.3% ahead.
Aug 27 - Sep 3 · week one
$1.09 vs $0.78
+39.6%
Adverge is at its working level from the opening day of the window and does not move off it again.
Sep 4 - Sep 10 · week two
$1.09 vs $0.73
+49.1%
Identical Adverge figure, to the cent, across a week where site sessions swung from 39,000 to 57,000.
Sep 11 - Sep 17 · week three
$1.09 vs $0.67
+63.7%
Adverge flat again. The margin widens only because the prior setup keeps sliding, and we would rather say so than claim acceleration.
Same traffic, same days. Two different results.
A 50/50 split is the only clean way to run this. Both stacks saw the same audience, the same seasonality and the same weekends. 1.04 million sessions, halved. Nothing here is modelled or normalized after the fact.
How to read this test.
Sessions and pageviews are the blog’s own GA figures for the whole site, and the split was 50/50, so each side is measured against half of them. Revenue per 1,000 sessions is the comparison that matters: it holds traffic constant across a window where daily sessions ranged from 39,000 to 57,000.
Figures are the blog’s own reported daily totals for Aug 27 - Sep 17, 2026 (22 days), from GA and the two ad stacks. The revenue lift (+49.4%) is computed on period totals, not on an average of daily percentages. That average is higher at +51.5%, because the smallest traffic days carry the largest percentages. Impressions (×1.30) and eCPM (×1.15) are period totals and multiply out to the revenue figure. The window opens on Aug 27; onboarding days before that date are not part of this account. Neither the blog nor the incumbent platform is named here.
Run the same test on your traffic.
Adverge tests head-to-head against your incumbent on live inventory, split 50/50 and measured daily. If your sessions are worth more than your stack is collecting, a month of side-by-side data will show it.