In online advertising, real-time bidding refers to the buying and selling of online ad impressions through real-time auctions that occur in the time it takes a webpage to load. Those auctions are often facilitated by ad exchanges or supply-side platforms (SSP).
As an ad impression loads in a user’s Web browser, information about the page it is on and the user viewing it is passed to an ad exchange, which auctions it off to the advertiser willing to pay the highest price for it. The winning bidder’s ad is then loaded into the webpage nearly instantly; the whole process takes just milliseconds to complete.
Teads SSP processes billions of auctions per day. Only a part of these auctions would lead to an impression, as a part of the inventory is deemed not interesting by the buyers.
The processing of these auctions has a significant computational cost. It involves some CPU-heavy computations, then the bid requests are sent to potential buyers, thus incurring some networking costs.
Project goal
The goal of this prediction model is to predict which auctions would lead to a sale and which auctions would not. By doing so, we can choose to not conduct auctions if we are confident that they would not lead to a sale. This can help us significantly decrease our infrastructure costs.
Before each auction, we dispose of some information about the context. This information can be used as an input to a classification model. The output would be wether this auction would lead to a sale or not.