Prebid.js Header Bidding Built Around a Timeout Budget

A wrapper is worth what it earns after latency has been paid for, and latency in this region is not what a European reference configuration assumes. On a median Jakarta or Manila mobile session, the page has already spent something close to two and a half seconds before an ad slot could sensibly call an auction. A 3000ms bidder timeout copied from a template built for German desktop traffic hands back a large share of the header bidding gain in abandoned sessions and unrendered first impressions. We build Prebid.js to a timeout budget derived from your own field data, and we fund new bidders by removing old ones. Choosing the bidder set: We rank the adapters you already call on three numbers read together: bid rate, win rate at the price they actually clear, and response time at p75 by geography. An adapter that answers late and wins a fraction of a percent is not free inventory coverage; it is paid for by every reader waiting on the auction. On a typical takeover we remove three to five adapters in the first month, the wrapper gets faster, and revenue does not fall. That outcome is dull and repeatable, and it is why we do it before we discuss adding anyone. Configuration decisions we will argue with you about: Price granularity buckets are cut to match your Unified Pricing Rules, not left on dense, so that the line item count stays inside what your ad ops team can maintain by hand. sendAllBids is off unless you can name the report that needs it, because the alternative is a targeting key explosion against Ad Manager's key limits. The currency module runs against a conversion file refreshed daily for SGD, MYR, IDR, THB and PHP, since a floor set in USD against a bidder responding in local currency is a floor that moves every morning. User ID modules are chosen on measured match rate in your markets rather than on the vendor's global average, and the ones that add a network call for nothing get cut. Consent runs through the TCF and GPP modules with the enforcement actually applied, not with enforcement disabled to make a warning go away. Measurement: An analytics adapter writes auction-level events to an endpoint you own and lands them in BigQuery next to your Ad Manager Data Transfer tables. Timeout rate, bid rate and win rate by adapter, geography and ad slot become a query rather than a quarterly slide from a partner with an interest in the answer. What we will not do: We will not add an adapter because a sales representative offered a bonus CPM for ninety days. We will not run refresh below a viewability gate, on a timer alone, or on units the reader never brings into view. We will not carry a bidder count the timeout budget cannot pay for, and if you want more demand than that, the honest answer is server-side, which is a different piece of work. We do not sell a demand relationship of our own, so there is no adapter we are trying to keep in your configuration. Handover: The configuration ships as a git repository you own, with the build pipeline, the module list pinned to a Prebid version, and a written note against every non-default setting explaining why it is not the default. Your engineers can read it without us in the room, which is the point. Not for: Apps monetised only through AdMob mediation. Sites where a JavaScript change takes a two-week release train, since the first four weeks are iterative. Publishers below roughly 15 million monthly ad requests, where the arithmetic does not clear our fee. Publishers who want the wrapper measured on CPM alone; we report session-level revenue, because CPM rises pleasantly when you lose the impressions that were never worth much.