[2F4-GS-10h] AI応用：経営と経済

[2F4-GS-10h-02] Estimating advertising wearout effects and causal effect estimation for optimal allocation using large scale RCT dataset

Using program promotion data in ABEMA Streaming service

〇Ryotaro Fujimoto^1,3, Yoshinao Umeda², Kenya Sakka², Takahiro Hoshino^1,3 (1. Keio University, 2. AbemaTV, Inc., 3. RIKEN)

Keywords:Advertising effect, Causal effect, heterogeneity

Maximizing the effectiveness of advertising under limitation of advertisement budgets is important in marketing. The concept of effective frequency, which takes into account "boredom" and "simple contact effects" that may occur when individuals expose advertisements several times, has been considered in advertising research. However, it is difficult to do randomized controlled trials in advertising. In this study, we analyze the determinants of the optimal number of ad contacts using randomly assigned ad contact data. We also clarify the heterogeneity of causal effects of ad contacts using machine learning approach. Specifically, we obtained the user's ad contact data obtained from ABEMA, a video distribution service provided by Abema TV Corporation, program viewing data and information on user attributes such as age and gender for analysis to model heterogeneity of causal advertising effects.