Sponsored Search and the Exploration-Exploitation-Outsourcing Dilemma

Motivated by the recently proposed (and eventually abandoned) partnership in sponsored search advertising between Yahoo! and Google, we examine a model for ad selection in which an agent must balance directly learning which ads perform best and outsourcing ad slots to another party in exchange for a share of any click revenue. Our framework captures a fundamental tension between the potential shortterm revenue gains from outsourcing, and the longer-term impacts on revenue due to the outsourced learning opportunities. Our main result shows that for broad classes of distributions over the arriving keyphrases (including the heavytailed distributions consistent with empirical search data), an outsourcing policy optimized for near-term revenue maximization always results in lower long-term revenue than a pure learning, no-outsourcing approach, and on a time scale not much longer than the outsourcing period itself.