Moving a Publisher Off AdSense Into Header Bidding

A publisher clearing only AdSense or AdX is taking one buyer's price. Putting Prebid.js in front of Google Ad Manager sets six to nine bidders against the same impression, and on Brazilian and Latin American traffic the display session RPM we have measured 90 days after cutover has landed between 20 and 55 percent above the AdSense baseline. Week one usually prints below the baseline. If the number has to go up inside a fortnight, start this in a different month. What gets built: Google Ad Manager as the ad server, Prebid.js on the page, Open Bidding for demand that will only transact server-side inside Google, and AdSense kept as backfill rather than removed. Line items are generated against a price granularity derived from your own clearing histogram, not Prebid's dense preset. Brazilian display clears low: on most sites we take on, more than half of winning bids land under R$3 CPM, so the buckets have to be fine where the money is and coarse above R$15 where almost nothing transacts. That single decision moves more revenue than any adapter added later. The order of work: Two weeks of reading before anything ships. GAM Data Transfer files, or the AdSense report broken out by ad unit, country, device and browser, so the cutover has a baseline that survives an argument in month three. Then a parallel run: Prebid live on a randomised slice of sessions, AdSense untouched on the rest, for at least 14 days including two weekends. We compare revenue per thousand sessions. CPM is not the measure, because CPM rises when you sell fewer impressions worse. Consent and inventory hygiene: A Google-certified CMP wired to TCF v2.2 where you have European traffic and to LGPD practice for Brazil, plus ads.txt and sellers.json checked line by line against each partner's own published file rather than pasted from an onboarding email. Roughly a third of the sites that arrive here believing their file is clean are carrying reseller lines nobody can account for. What we will not do: We will not launch with a long adapter list. Five to seven at cutover, and each addition afterwards earns its place against a two-week holdout of the stack without it. Past the eighth adapter most are net negative once timeout loss and page latency are counted. We will not install a wrapper whose auction we cannot read; if the analytics adapter cannot write to your BigQuery, we do not run it. And we will not add ad slots to move the number. Density is your editorial decision, and we will tell you what each new slot costs in p75 INP before you make it. Not included: Direct sold campaigns and rate cards, refresh and lazy load policy, adblock recovery. Each is separate work and each is on this profile. Video and CTV are out of scope; this practice does display and native on web. Who this is not for: Sites under about 300,000 monthly sessions. Exchange minimums and the engineering overhead do not pay back at that size, and better placements on AdSense is the honest answer. Sites with an open invalid traffic finding, which should be resolved first because no exchange will take the supply anyway. Publishers whose audience is mostly in-app webview from one social source, where auction latency eats the gain. And anyone wanting the migration compressed into two weeks. We have done it that way and it produces a stack nobody can debug when a partner breaks at 2am on a Saturday. Everything stays yours. Ad Manager in your account, Prebid config in your repository, line items generated by a script your engineer can run without us.