The (Marginal) Value of a Search Ad: An Online Causal Framework for Repeated Second-price Auctions
New causal framework shows auto-bidders overpay by not accounting for organic traffic value.
Researchers Yuxiao Wen, Zihao Hu, Yanjun Han, Yuan Yao, and Zhengyuan Zhou have published a paper titled "The (Marginal) Value of a Search Ad: An Online Causal Framework for Repeated Second-price Auctions" set to appear at ICML 2026. The work addresses a fundamental flaw in digital advertising: existing auto-bidding systems assign value to an ad slot based solely on the revenue generated when the ad is shown or clicked. This ignores the fact that advertisers can still generate sales organically (e.g., via Google or Amazon organic search results) even if they lose the auction. The true value of winning an ad slot is the marginal gain — the treatment effect of paid exposure over organic visibility.
The authors formalize this as an online learning problem in second-price (Vickrey) auctions under a causal perspective. They develop bidding algorithms that achieve rate-optimal regret under several feedback models, including scenarios with full feedback and bandit feedback. A key innovation is exploiting the information revealed by the second-price payment rule (the price paid is the second highest bid), which strictly improves regret bounds compared to analogous problems in first-price auctions. This means advertisers can learn to bid more efficiently over time, reducing overspend on auctions where organic results would have sufficed. The results have direct implications for platforms like Google Ads, Amazon Marketplace, and other search advertising ecosystems.
- Models ad value as a causal treatment effect (outcome difference between winning and losing auction), not just revenue from clicks.
- Achieves rate-optimal regret in repeated second-price auctions, with algorithms using the second-price payment rule to outperform first-price auction baselines.
- Reduces wasteful ad spend by accounting for organic search traffic, addressing a key inefficiency in auto-bidding systems on Google Ads and Amazon.
Why It Matters
Could save advertisers billions by eliminating overbidding on searches where organic results already convert.