Designing Ad Auctions with Private Constraints for Automated Bidding

In online advertising, automated bidding (auto-bidding) has become a widely-used tool for advertisers to automatically make bids on different impressions in real time. Instead of submitting bids for each single impression, advertisers in auto-bidding submit their high-level objectives and constraints to the auto-bidding tool, and observe the cumulative advertising performances after all the auctions within a time period have been finished. Motivated by the features of automated bidding, we aim to design auctions with private financial constraints for value-maximizing bidders. Specifically, we consider budget and ROI, the two most common financial constraints in online advertising, as the private information of advertisers, and analyse the conditions of truthfulness. We show that every non-decreasing function with budget as input could be mapped to a truthful auction mechanism with budget and ROI as input, but this mapping procedure also introduces complex value grouping structures into mechanism design. To achieve feasible and implementable auctions, we design a truthful auto-bidding auction mechanism with adjustable rank score functions. As the key design to guarantee truthfulness, our auction utilizes the bidder's budget constraint to compute a critical ROI, which enables comparisons between the budget and ROI constraint. We conduct experiments under different auto-bidding settings to validate the performance of our proposed auction in terms of revenue and social welfare.