chat & UGC platform × Adverge · 26-day A/B test vs. the prior setup
26/26
days ahead. Every day of the test, on identical traffic.
A high-volume chat and user-generated content platform ran Adverge against its incumbent setup for 26 consecutive days — 100,000 sessions a day to each side.
Revenue per 1,000 sessions went from $15.08 to $17.38. The daily margin averaged +16.7%, and never once went negative.
Strip out the two days the prior setup’s delivery failed and the honest steady-state number is +13.1%.
Revenue per 1,000 sessions
$15.08 → $17.38
Feb 25 – Mar 22, 2026
Steady state, outliers removed
+13.1%
Mar 16–17 excluded — the number we would defend
Total revenue, period
+15.3%
on period totals rather than a daily average
Twenty-six days, side by side.
Revenue per 1,000 sessions. Both stacks normalized to 100,000 sessions per day.
Sessions were held identical on both sides, so revenue, revenue per session and session RPM all move by the same percentage — one line carries the whole story. The shaded band is Mar 16–17, when the prior setup’s delivery fell away. Those two days lift the daily average to +16.7%; with them removed it reads +13.1%, which is the number we quote.
Daily numbers — all 26 days
| Date | Adverge / 1k sessions | Prior setup / 1k sessions | Difference |
|---|---|---|---|
| Feb 25, 2026 | $16.37 | $14.87 | +10.07% |
| Feb 26, 2026 | $18.73 | $15.98 | +17.23% |
| Feb 27, 2026 | $17.39 | $15.00 | +15.91% |
| Feb 28, 2026 | $17.66 | $15.85 | +11.43% |
| Mar 1, 2026 | $17.15 | $15.90 | +7.89% |
| Mar 2, 2026 | $16.17 | $14.55 | +11.08% |
| Mar 3, 2026 | $15.39 | $13.74 | +12.04% |
| Mar 4, 2026 | $14.87 | $13.19 | +12.72% |
| Mar 5, 2026 | $15.60 | $13.26 | +17.63% |
| Mar 6, 2026 | $17.33 | $14.77 | +17.26% |
| Mar 7, 2026 | $17.78 | $15.87 | +12.03% |
| Mar 8, 2026 | $18.36 | $16.09 | +14.07% |
| Mar 9, 2026 | $17.68 | $15.80 | +11.94% |
| Mar 10, 2026 | $17.09 | $15.17 | +12.61% |
| Mar 11, 2026 | $18.26 | $16.21 | +12.65% |
| Mar 12, 2026 | $18.64 | $16.00 | +16.50% |
| Mar 13, 2026 | $18.22 | $16.49 | +10.52% |
| Mar 14, 2026 | $18.20 | $16.00 | +13.77% |
| Mar 15, 2026 | $17.55 | $15.33 | +14.49% |
| Mar 16, 2026 | $14.35 | $9.73 | +47.45% |
| Mar 17, 2026 | $15.14 | $8.86 | +70.76% |
| Mar 18, 2026 | $17.21 | $15.51 | +10.94% |
| Mar 19, 2026 | $19.05 | $16.88 | +12.81% |
| Mar 20, 2026 | $19.69 | $17.40 | +13.13% |
| Mar 21, 2026 | $19.25 | $17.03 | +13.06% |
| Mar 22, 2026 | $18.68 | $16.50 | +13.22% |
26 days.
26 wins.
Each bar is one day’s revenue difference against the prior setup. Not a monthly average hiding bad weeks — every individual day came out ahead.
Bars are clipped at +25%. Two exceed it — Mar 16 (+47%) and Mar 17 (+71%) — the two days the prior setup’s delivery fell away. Those are the bars we discount, not the ones we quote.
Feb 25 – Mar 5 · opening
$16.59 vs $14.70
+12.8%
Geo-aware floors and curated demand take hold. Adverge is ahead from the first day of the test.
Mar 6 – Mar 15 · middle
$17.91 vs $15.77
+13.6%
Both sides rise together. The gap holds across weekday and weekend traffic alike.
Mar 18 – Mar 22 · close
$18.77 vs $16.66
+12.7%
The strongest five days of the test for both stacks, with the margin holding to the decimal. Adverge peaked at $19.69 on Mar 20.
Three windows, three readings within half a point of each other. The lift is not an artifact of when you start counting.
What we actually changed.
Same inventory. Same sessions. Four changes to how demand was called.
01
First-party data sync
Logged-in and returning-user signals from the platform’s own graph were synced to demand partners before the auction, not inferred inside it. Buyers bid on a known audience instead of an anonymous one.
02
AI price floors, throttled by geo
A chat platform’s audience mix shifts by the hour as regions wake up. Floors were modelled per geo and re-set as the mix moved, so high-value traffic was never priced off a global average and low-value traffic was never priced out of the auction entirely.
03
Demand curation for UGC
User-generated pages are a hard sell for brand buyers. We curated the demand set deliberately — who is invited, on which content, at what price — instead of opening every placement to everyone and letting blocklists sort it out after the fact.
04
Refresh built for long sessions
Time on page is high here and pageviews are low; an impression-first stack undercounts the session badly. Refresh fired on observed engagement rather than a timer, so the platform monetized the visit instead of the pageview.
The traffic never changed. The read on it did.
No new inventory, no layout change, no added placements. Both sides were handed 100,000 sessions a day and the same open demand. The difference is what each stack knew about the session before it asked for a bid.
Why it worked here.
Chat and UGC inventory breaks the assumptions most stacks are tuned on. Sessions run long, pageviews stay flat, the audience rotates through geos all day, and the content is written by users rather than an editorial desk. A stack that prices on a global floor and refreshes on a clock is guessing at all four. Reading the session — who is here, where from, how engaged, on what page — is what moved the number.
Figures are the publisher’s own reported daily totals for Feb 25 – Mar 22, 2026 (26 days), normalized to 100,000 sessions per day on both sides. Revenue per 1,000 sessions averages $17.38 for Adverge against $15.08 for the prior setup; the daily difference averages +16.7%, period totals give +15.3%, and removing Mar 16–17 — two days on which the prior setup’s delivery dropped sharply — gives +13.1%. Neither the publisher 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, measured daily. If your sessions are worth more than your stack is collecting, a month of side-by-side data will show it.