AI, Machine Learning and Attribution: Practical Application in Marketing at IAG
A Research Wonks semi-exclusive, this deck is from a major brand advertiser, IAG, sharing powerful insights on how uplift modeling, with experiments, delivers greater confidence in what is working in their ad mix than propensity modeling with traditional observational methods.
Willem Paling, PhD, Director, Customer & Growth Analytics at IAG, is the author of the material. He agreed to let Research Wonks share the presentation.
Uplift modeling is one of the best, best practices in marketing, once the province of direct mailers but now practiced more broadly in digital marketing, too. When it comes to audience targeting, many marketers are making a critical mistake in relying on propensity modeling instead of uplift modeling.
Propensity models predict which users look likely to convert. The problem with this is the advertiser pays to reach many people who would have converted regardless of the ads. Those are the “sure things,” in the vernacular of uplift modeling. With uplift modeling, the advertiser instead tries to target only those most susceptible to brand persuasion, “the persuadables,” while avoiding the “sure things” and the “do-not-disturbs.”
The deck explores some of the advertising experiments IAG has undertaken on paid search and other media and how they’ve applied their conclusions, pairing measurement and optimization in a way that other marketers could learn from.