a global chat platform × Adverge · 36-day A/B test
3.1×
the revenue, on the same traffic.
A high-volume chat platform split its traffic against its incumbent monetization partner and let both run for 36 consecutive days.
Same inventory. Same demand sources. Only the layer deciding how demand was called changed.
Adverge finished ahead on all 36 days.
Revenue
+214%
vs. the incumbent, normalized to the same traffic
Revenue per session
$4.37 → $13.70
what a single visit is worth
Page RPM
$1.27 → $3.97
what a thousand pages are worth
36 days.
36 wins.
Not a monthly average hiding bad weeks. Every individual day of the test came out ahead.
The same lift, three ways.
Incumbent
Revenue
Revenue per session (RPS)
Page RPM
Both platforms are measured against the same sessions and the same pageviews, so revenue, RPS and page RPM move by the identical percentage. The three views answer three different questions: what came in, what a visit was worth, and what a page was worth.
What we actually changed.
01
Refresh optimization
Driven by first-party data signals and observed user behavior rather than a fixed timer. On a chat platform sessions run long, so when a refresh fires matters more than how often.
02
Timeout optimization
Tuned so slower, higher-paying bidders still land instead of being cut off. A timeout set for speed quietly discards the bids worth the most.
03
Demand orchestration
The same partners, called in a better order and with better information about the session they were bidding into.
04
Demand throttling
Low-value, high-noise demand held back so it stopped crowding the auction and slowing every other bidder down.
Same demand sources. The bidders never changed.
No new partners were added for this test. The entire difference came from how the auction was run: when it fired, how long it waited, who was asked, and in what order.
Why it worked here.
A chat platform is the hardest kind of inventory to monetize on a timer. Sessions are long, attention is uneven, and pageviews do not map cleanly onto engagement. A stack that refreshes on a clock is guessing. A stack that refreshes on behavior is reading the session, and gets a real impression at the moment it is worth something.
Figures are the publisher’s own reported totals across the 36-day test, normalized so both platforms are measured against the same traffic. The publisher and the incumbent platform are not named here at the publisher’s request, and absolute revenue figures are withheld for the same reason.
Run the same test on your traffic.
Adverge tests head-to-head against an incumbent on live inventory, measured daily. If your sessions are worth more than your stack is currently collecting, a month of side-by-side data will show it.