Mile

Prebid.js 11.9.0 Released

Prebid.js published release 11.9.0 on 2026-04-29 (tagged via GitHub Actions). The release adds new features including an Encypher RTD provider that surfaces C2PA content provenance signals, multiple new bid adapters (goadserver, mtc, tne_catalyst), and a pgamdirect analytics adapter. It also introduces an option in GPP control modules to further restrict activities. The update contains a core bug fix limiting mergeConfig notifications to updated topics and several maintenance changes: disabling user syncs in Firefox and WebKit-based Chrome, build-system migrations (native Gulp v5 sourcemap handling, webpack source-map changes), adapter improvements (GumGum, Vidazoo, Mile, Showheroes, Limelight), and dependency bumps. The release is relevant to publishers, header-bidding integrators and ad-tech implementers managing Prebid.js builds and privacy/consent flows.

Empowering Publishers with Independent AI for Yield Boost

Vijay Kumar, founder and CEO of Mile, argues that publishers lack the neutral, production-grade ML infrastructure required for real-time yield optimisation. Exchanges and SSPs optimise for marketplace health and GMV, not publisher RPM, creating a "last mile" gap where publishers control final auction settings but lack impression-level decisioning. Effective publisher-side optimisation requires impression-level logging across multiple dimensions, feature engineering, daily model retraining, sub-50ms inference in the ad request path, and always-on A/B testing (typically 80/20 control splits). Independent providers can deliver this infrastructure while preserving publisher data ownership and strategy. Kumar notes practical tradeoffs — dynamic floors often reduce fill by 3–6pp, models require 2–4 week cold-starts, and the practical publisher scale sweet spot is ~5–10M impressions/month per property.

Maximize Revenue: Optimize Your Programmatic Floor Pricing Strategy

The article explains how publishers should measure and optimize programmatic floor pricing to close the gap between static floors and real-time auction pricing. It argues static floor tables based on historical averages often misalign with DSPs' per-impression repricing, producing thinner auctions, fewer bidders, and lost revenue. The author recommends using holistic RPM (total programmatic revenue per thousand ad opportunities) as the primary success metric, supplemented by CPM, fill rate, and bid density for diagnostics. Best practices include running floored and control traffic in parallel (e.g., 10–20% control), analyzing bidder participation by placement/device/geography, and signaling floors into auctions rather than enforcing them only at the ad server. Proper reporting granularity and iterative tests can raise yield without harming fill.