Fairness in the Autobidding World with Machine-learned Advice

Online advertising platforms have commonly focused on maximizing total advertiser value (or welfare) to attract advertiser traffic and spend, and have resorted to machine learning predictions on advertiser values (also known as machine-learned advice) to improve ad auction designs and thus total welfare of advertisers. Yet, such improvements could come at the cost of individual bidders' welfare, consequently eroding fairness of ad platforms, and do not shed light on how particular advertiser bidding strategies impact individual fairness. Motivated by this, we present a novel fairness metric that measures an individual bidder's welfare loss, and also uncovers how advertiser strategies relate to such losses. Under this metric, we then study how ad platforms can utilize ML advice to improve welfare guarantees and fairness on the individual bidder level. We first motivate a simple approach that directly sets such ML advice as personalized reserve prices when the platform consists of \textit{autobidders} who maximize value while respecting a return-on-ad spent (ROAS) constraint. Under parallel VCG auctions with such ML advice-based reserves, we utilize our fairness metric to present a worst-case welfare lower-bound guarantee for an individual autobidder, and show that platform fairness is positively correlated with ML advice quality as well the scale of bids induced by the autobidder's bidding strategies. Further, we prove an impossibility result showing that no truthful, and possibly randomized mechanism with anonymous allocations can achieve universally better fairness guarantees than VCG, in presence of personalized reserves based on ML-advice of equal quality. Finally, we extend our fairness guarantees to generalized first price (GFP) and generalized second price (GSP) auctions.