Generalized Agent-mediated Procurement Auctions

Procurement auctions (where the auctioneer needs a service and bidders offer it at their own conditions) are an appealing method for on-line service selection. They can improve service features and cost by exploiting the competition between different service providers. Software agents, acting on behalf of human users and organizations, are essential in making such auctions practical and usable. Since conveying user preferences to the agents in a faithful and complete way is virtually impossible, we advocate an approximate approach, where only partial preferences are formalized, and users pick their choice from a short list of options selected by the agents by means of those partial preferences. Another peculiarity of our scenarios is that there may be no contracts with null utility for a given bidder. These features affect the classical, desirable properties of standard auction mechanisms. We prove some impossibility results concerning truthfulness and (a qualitative analogue of) revenue. Then, we investigate a novel auction mechanism that is "almost" truthful in the sense that any strategic deviation from truthfulness has limited impact on the auctioneer's revenue.