HOB bidding strategy boosts GMV 3% across heterogeneous ad channels

Optimizing advertising campaigns across multiple channels is notoriously difficult because each channel uses different auction mechanisms—first-price vs. second-price pricing, eCPM vs. UE-augmented ranking, uniform vs. non-uniform bidding. Advertisers also have shared campaign-level constraints like budget and ROAS. Existing strategies fail to align marginal costs (MC) across such heterogeneous environments, especially in first-price auctions with organic results influencing paid slots.

Enter HOB (Holistically Optimized Bidding), a framework that redefines marginal cost computation to work uniformly across all channel types. At the global level, HOB derives channel-specific MC formulas and coordinates them via a shared MC target. At the local level, it models free-win probability and winning-price uncertainty using a zero-inflated exponential distribution, enabling efficient surplus-optimal bids in non-uniform first-price auctions. The team proved mathematically that any interior optimum satisfies marginal cost equalization across channels.

Experimental validation was rigorous: offline benchmarks, industrial log replay, and large-scale online A/B tests on a commercial DSP. Results show a consistent 3.0% lift in Gross Merchandise Value (GMV) while strictly adhering to return on ad spend (ROAS) constraints. The paper (v2, June 2026) includes a substantially revised MC framework and expanded experiments. This work has immediate practical value for programmatic advertising platforms, offering a deployable algorithm that handles real-world auction diversity without sacrificing performance.