This talk presents a work in progress that extends our earlier research on optimization in on-line advertising. One of the dominant methods for filling available advertisement slots on-demand is by using a waterfall mechanism. In a waterfall mechanism advertising opportunities are provided using a function which will select the next pre-screened advertiser to ask to fill an available ad slot, given a current system state. The function can take into account various candidate metrics, for example revenue and historical response times. In this highly time sensitive problem, ads should be filled within 200 milliseconds. We developed effective metaheuristics and focused real-time dynamic programming based solutions (having bounded sub-optimality) for two different variations of that problem.
We are now turning our attention to other contracting problems namely procurement in spot markets. Spot markets are important in logistics and supply chain management contracting and also in electricity and natural gas contracting and many other applications. We hypothesize that in logistics contracting, the existence of multiple spot markets with duplicate supply and demand postings, makes estimating the true supply and demand difficult. We suspect that one-by-one solicitation will be an attractive alternative to auctions, since full information on demand complexities are not always effectively expressed, and because both demand or supply could be covered or contracted before the auctions are completed. In this joint work with Dmitri Arkhipov, Michael Dillencourt, Roger Lloret-Batlle, and John Turner, we explore the potential of these specialized mechanisms in spot markets.