Getting the pace right: Performance of budget allocation heuristics in online advertising
比较了三种在线广告预算分配启发式方法（均匀、修正均匀、水位线）的表现，发现水位线法优于其他方法，并提出了改进的自适应水位线法。
Online advertisers use pacing heuristics to allocate their ad budget over time: Even Pacing aims to achieve an equal budget spent across hours; Modified Even Pacing does the same but excludes specific periods; and Waterlevel Pacing spends the budget according to traffic patterns. This article compares the performance of these pacing heuristics. Specifically, we run a counterfactual study using three campaign datasets and then benchmark the profitability of each of these heuristics against that of a profit-maximizing pacing model. Despite being the default setting in most campaign management systems, Even Pacing consistently underperforms compared to other heuristics, and so does Modified Even Pacing. Waterlevel Pacing outperforms other heuristics and even a profit-maximizing pacing model under certain conditions. These findings lead us to develop a new pacing heuristic by further improving Waterlevel Pacing (called Adaptive-Waterlevel Pacing). To examine the conditions under which these pacing heuristics perform best, we systematically vary the factors that may impact their performance in a simulation study, specifically accounting for advertisers’ different campaign goals (i.e., reach, profit, or a mix of both). Our analyses reveal that Waterlevel Pacing outperforms existing industry heuristics in the presence of uncertainty, with Adaptive Waterlevel Pacing offering further improvements when the goal is to generate profit.