What’s your project about?

My research is about budget allocation in online (display) advertising. Budget allocation decisions are at the core of marketing and have a long tradition in the marketing literature.

What’s the research problem?

A dimension about the budget allocation decision that is unique to online advertising is that advertisers need to decide how to allocate their budget over the hours of a day. In practice, they use so-called pacing heuristics to do that. They are easy-to-implement, but—compared to solutions proposed in the literature—not profit-maximizing for the advertiser. Those solutions suggested in the literature are, on the other hand, hard to implement and require accurate prediction of various input metrics. We study under what circumstances pacing heuristics used in practice come close to profits generated by profit-maximizing pacing.

Which solution does your paper bring to the problem?

Specifically, we argue that being unable to accurately predict required input metrics for profit-maximizing pacing makes pacing heuristics attractive, as their profit does not deviate too much from that being generated by profit-maximizing pacing anymore. 

How did you study this?

My empirical study is based on two analyses: In the first analysis, We run a counterfactual analysis on three different industry datasets. Here, we forecast input metrics for profit-maximizing pacing and calculate its profits when those forecasts are (a) 100% accurate vs (b) based on simple prediction models that come with forecasting errors. We then compare those profits to profits that would have been generated had the advertiser used simple pacing heuristics (Even Pacing and Waterlevel Pacing). In the second analysis, we run a simulation study, in which we vary the forecasting error on my predictions to see what level of forecasting error would result in those simple pacing heuristics becoming attractive for the advertiser (because their profits now come close to those from profit-maximizing pacing).

What did you find?

We find that one simple pacing heuristic performs particularly well. It is called Waterlevel Pacing and the budget is spent according to traffic fluctuations over a day. 

What can practitioners learn from these results?

Advertisers will know when to use which pacing heuristic and how much money they are leaving on the table by using them (vs profit-maximizing pacing). Advertising platforms offering different pacing heuristics in their systems will learn that the standard option they provide (Even Pacing) is not the best pacing heuristic; they should add Waterlevel Pacing to their systems and might even make this the default option in their campaign management systems.