Prebid and Mediation Stack Rebuild for Publishers

The stacks we open usually carry between 19 and 30 demand partners, and six of them produce around 80% of net programmatic revenue. Removing most of the rest raises revenue rather than lowering it, because every additional bidder adds latency to the page, dilutes the auction it joined, and takes a fee out of the same dollar. How the diagnosis runs: We instrument the wrapper before touching it. A Prebid analytics adapter reports bid rate, win rate, timeout rate and response time per bidder, joined to ad server revenue at ad unit level. The figure that decides a partner's future is net revenue per thousand bid requests after latency cost, not gross CPM, which is the number partners quote and which flatters resellers. A bidder clearing under 1.5% of net programmatic revenue across 60 days while adding more than 100ms at the 95th percentile comes out of the stack. Floors and pricing rules: Floors are set per ad unit, geography and device from the historical clearing distribution, not as one number applied everywhere. In Google Ad Manager that means unified pricing rules that do not quietly contradict the Prebid floors module, a conflict present in most accounts we open and usually costing money on the units that matter most. Price granularity is matched to the real bid distribution instead of the default buckets. Supply path: Buyers now optimise for the shortest path to your inventory. We clean ads.txt and sellers.json so the direct lines are unambiguous, remove reseller entries duplicating a relationship you already hold, and take the resulting map into the conversation with the desks that matter for your audience. Fewer and cleaner paths is a commercial argument you can make to a buyer, not only hygiene. Apps: Where there is an app, mediation is rebuilt on the same logic. Bidding replaces waterfall instances wherever the network supports it, per-instance floors come from observed eCPM rather than the template, and placement frequency is reviewed against session length instead of pushed to whatever the SDK permits. Consent and region: Fill collapses without a valid consent signal. We check the CMP against TCF v2.2 for European traffic and confirm the string actually reaches the bidders, and we review what Saudi Arabia's PDPL and the UAE data protection law ask of a publisher serving Gulf audiences with international demand attached. Page experience: Ad-driven layout shift and input delay are treated as revenue problems, not engineering complaints. Slot reservations to hold layout, below-fold loading tied to viewport distance, and a refresh policy at 30 seconds of continuous viewability with a per-session cap. We do not implement refresh below 20 seconds, or on slots that were never viewable, whatever the uplift looks like in the spreadsheet. Not included: Daily ad operations and trafficking. Selling direct campaigns. CMS or template rebuilds beyond the ad slots themselves. Video player selection, unless video is the reason you called. What we refuse: Made-for-advertising layouts. Ad density past the agreed policy, including injected units deep in infinite scroll. A wrapper build whose source we cannot read. And we take no share of media. The fee is the fee, so advice about removing a partner is never advice about our own revenue. Who this is not for: Sites whose traffic is bought, where the arithmetic only closes at a density we will not implement. Publishers who want a partner added rather than a stack examined, and publishers whose programmatic revenue is small enough that the honest recommendation is to spend the money on direct sales instead.