Maximizing net income of the auction waterfall with an abort decision tree

TL;DR

This paper develops an abort decision tree to skip unlikely auctions in an online ad waterfall, maximizing net income by reducing transaction costs based on auction and context features.

Contribution

It introduces a novel decision tree-based abort rule that improves net income in ad auctions by selectively skipping low-probability auctions.

Findings

Abort decision tree outperforms full waterfall in net income.

Higher transaction costs lead to greater savings and net income gains.

Using context features improves abort decision accuracy.

Abstract

An online auction waterfall for an ad impression may contain auctions that are unlikely to result in a winning bid. Instead of always running through the full auction sequence, one could reduce the transaction cost by predicting and skipping these auctions. In this paper, we derive the auction abort rule that maximizes the net income of the waterfall under certain conditions, knowing only the publisher tag of the current auction and the ad request context. The net income is defined as the payoff (revenue) minus the transaction cost. We translate the abort rule into a purity measure and propose a corresponding split criterion for a decision tree. Training and testing on randomly sampled data indicate that the abort decision tree performs better than the full waterfall and the abort rule that makes use of only the publisher tag feature. When the transaction cost is higher, the cost saving, and thus net income gain, is higher for either abort decision rule.