A web publisher lifted programmatic revenue 31% in 60 days with AI floors
Example scenario. This page shows how we approach this kind of publisher and the results we aim for. It is not a report on a named customer.
The challenge
A high-traffic content network ran static floors set two years earlier and a bloated 14-bidder client-side wrapper. CPMs had eroded under first-price bid shading, pages were slow, and the team had no holdout measurement to know what was actually working.
What we did
- Connected both GAM networks to the PubMonetX console and ingested 30 days of auction logs to build a bid-landscape baseline.
- Activated AI floors in shadow mode across geo × device × ad unit segments, then promoted rules that beat the 5% holdout for 14 consecutive days.
- Pruned the wrapper from 14 to 8 client-side bidders, moved the long tail to Prebid Server, and tuned timeouts per geo.
- Rebuilt the article template's slot reservation CSS, eliminating ad-driven CLS entirely.
How this was measured
On a live engagement we measure results against a randomized holdout: a small share of traffic keeps the old setup while the rest runs the new one, and the uplift is the difference between the two groups over the period. Comparing against a control group, not against last month, keeps seasonality and traffic changes out of the number.
What would this look like on your site?
Start with the free audit. We'll show you where your current setup is losing money.
Get the free audit